Building Ontologies for Knowledge Management Applications in Group Sessions

Ontological Representation for Learning ObjectsJian Qin and Christina FinneranSchool of Information StudiesSyracuse UniversitySyracuse, NY 13244 USA+1 315 443 5642{jqin, cmfinner@}ABSTRACTMany of the existing metadata standards use content metadata elements that are coarse-grained representations of learning resources. These metadata standards limit users’ access to learning objects that may be at the component level. The authors discuss the need for component level access to learning resources and provide a conceptual framework of the knowledge representation of learning objects that would enable such access.KeywordsLearning objects, component access, intelligent access, knowledge schemas, metadata, ontologies INTRODUCTIONAs the design of search interfaces advances, digital libraries are witnessing limitations based on the underlying representation of the data. The describing author defines the granularity of a resource statically through using current metadata schemas [8]. The granularity of digital library resources will inevitably limit the user from retrieving finer-grain resources. Our paper will discuss an ontological approach to representing data within a digital library to enable more component level access.Learning objects refer to any entity, digital or non-digital, that can be used, re-used or referenced during technology-supported learning [9]. Broadly speaking, learning resources usually refer to documents or collections, whereas learning objects to the components of a document or collection. However, “learning objects” according to IMS [6] standards refers to any object, regardless of granularity.Within the domain of education digital libraries, one of the important goals of metadata is to enable the retrieval and adaptation of one learning object to another learning situation [1]. Providing the learning objects that the users seek, whatever the granularity, is the essence of contextual design [3] and essential for an effective digital library interface. Representations of learning resources must support the finest-grained level of granularity required by the core technologies as suggested in [7], in addition to application and support technologies. Objects within a learning resource need to be encoded in a way that they can be recognized, searched, referenced, and activated at different granularities. Other researchers have been addressing related technological solutions, such as dynamic metadata and automated component descriptions [2,8]. We will focus on the knowledge representation needs, and introduce a conceptual framework of an ontological approach toward metadata. Finally, we discuss how component-level representation contributes to user-focused interface design.ONTOLOGICAL APPROACH TO METADATAMetadata standards pose different levels of representation granularity, as demonstrated in Figure 1. Dublin Core (DC) [4] provides basic factual description, which is used most commonly in creating collection or resource level metadata. The educational extension of DC specifies contextual factors, such as the resource’s target audience and pedagogical goals. IEEE’s Learning Object Metadata (LOM)/IMS metadata standard defines more specific educational and technical parameters for learning resources [5,6]. These three metadata standards are best situated to represent learning resources at the collection or resource level. To reach a finer-grained level, where components in a resource are represented and correlated, knowledge schemas play an important role in in-depthrepresentation and more refined user access.Figure 1.Representation FrameworkLEAVE BLANK THE LAST 2.5 cm (1”) OF THE LEFTCOLUMN ON THE FIRST PAGE FOR THE COPYRIGHT NOTICE. Through an informal survey of the NSDL collections, we found that search, browsing, and navigation capabilities vary widely depending on the purpose, scope, and subject area of the collection. However, collection or documentlevel metadata dominates all types of searches available. A lack of finer-grained representation is becoming a crippling factor for user interfaces to provide in-depth searching for learning objects.SAMPLE MODEL FOR COMPONENT REPRESENTATIONAn ontological representation defines concepts and relationships up front. It sets the vocabulary, properties, and relationships for concepts, the result of which can be a set of rich schemas. The elements accumulate more meaning by the relationships they hold and the potential inferences that can be made by those relationships. The key advantage of an ontological representation within the realm of learning objects is its ability to handle different granularities. In order to describe learning resources at the collection level (e.g. web site) and further describe each of the components (e.g. interactive applet, image), relationships must be identified when the data are input. Only by having description at the component level will specific learning objects be able to be retrieved by users. Figure 2 demonstrates how even with a seemingly simple laboratory-learning object, fine-grained description of the component level can enable better access. For example, if an instructor is interested in a graph of steam gauge metrics, s/he should be able to search at the component level, rather than having to guess what type of resource(e.g. textbook, lab) might contain such a graph.An ontological model may be created based on the example in Figure 2. Each component in the model is normalized into a group of classes under class Lab. The attributes for Lab include “object subject,” “object URI,” and “parent source,” which are inherited by all its subclasses. The “object content” attribute is local to subclass Formula and also reused in other subclasses. A unique feature in this sample model is the reuse of classes in defining attribute types (e.g., the Hydrogeology class is reused in attribute objectSubject). This model can be converted directly into a Resource Description Framework (RDF) format, which can then be used as the motor behind intelligent navigation and retrieval interfaces. By creating ontologies for learning resources, we will be able to generate a set of knowledgeschemas for buildingknowledge bases andrepositories that can beshared and reused by systemdevelopers.Figure 3. A sample component representation model CONCLUSIONThe goal of education digital library interfaces is to support users, whether they be educators or learners, in accessing useful learning objects. The user will determinewhat is useful, and they also should be given the opportunity to search for components that may be useful.The ontological approach to representing learning objects provides a framework upon which to build more intelligent access within digital libraries.REFERENCES1. Conlan, O, Hockemeyer, C., Lefrere, P., Wade, V. andAlbert, D. Extending Educational Metadata Schemas toDescribe Adaptive Learning Resources. HT 2001,Aarhus, Denmark.2. Damiani, E., Fugini, M.G. and Bellettini, C. AHierarchy-Aware Approach to Faceted Classificationof Object-Oriented Components. ACM Transactions onSoftware Engineering and Methodology 8(3): 215-262,July 1999.3. Dong, A. and Agogino, A.M. Design Principles for theInformation Architecture of a SMET Education DigitalLibrary, JCDL 2001, June24-28, 2001, Roanoke, VA.4. Dublin Core Metadata Initiative 5. IEEE/LTSC. .6.IMS Learning Resource Metadata /metadata/7. Laleuf, J.R. and Spalter, A.M. A ComponentRepository for Learning Objects: A Progress Report.JCDL 2001, June24-28, 2001, Roanoke, VA.8. Saddik, A.E., Fischer, S. and Steinmetz, R. Reusabilityand Adaptability of Interactive Resources in Web-Based Educational Systems. ACM Journal ofEducational Resources in Computing 1(1), Spring2001.9. Wiley, D.A. Learning Object Design and SequencingTheory. Ph.D. Thesis. Brigham Young University,2000.。

Knowledge Engineering Ontologies, Constructivist Epistemology, Computer Rhetoric: A Trivium for the Knowledge Age.Arthur StuttKnowledge Media Institute, The Open University, UKA.Stutt@Abstract: In this (largely speculative) paper I want to suggest that certain ideas whichhave recently been developed by knowledge engineers can fruitfully be applied in theeducational context. I have two main reasons for thinking that this is so: firstly, ideas fromknowledge engineering have been applied in the past with some success; secondly, overthat last five years or so there has been a paradigm shift in knowledge engineering awayfrom a mining-and-transfer view of knowledge acquisition to a constructivist, modellingapproach which has broad similarities to currently prevalent ideas in educational theory.Briefly put, knowledge engineering can provide the conceptual and other tools which canfacilitate the construction of knowledge models. The relevant knowledge engineering ideasare illustrated with a knowledge model from the archaeology domain and the possiblebenefits for learners are outlined.1.IntroductionOur own Middle Ages, it has been said, will be an age of “permanent transition” for which newmethods of adjustment will have to be employed. [Eco 1987 p. 84]In the middle ages, students had to acquire skills in a set of three communicative arts (the trivium of grammar, dialectic and rhetoric) as well as studying the mathematical quadrivium (arithmetic, geometry, astronomy and music). In our postmodern age the variety of sources and types of knowledge has expanded considerably and is continually expanding. In this paper I want to suggest that recent developments in knowledge engineering have much to offer as the basis for a modern trivium which will enable learners to cope with this expansion. This trivium consists of ontological and other (problem solving) categories from knowledge engineering, a basically constructivist rather than mirroring approach to knowledge and, finally, computational models of and computer support for argumentational structures. I want to suggest that, while there are other possibilities for constructivist tools, any such tool intended to aid students in learning reasoning and problem solving skills would do well to incorporate recent knowledge engineering ideas. The paper outlines the use of knowledge engineering in education, introduces constructivist theory, discusses recent developments in knowledge engineering and illustrates these with a knowledge model from the archaeology domain. The benefits for learners and a possible tool are outlined.2.The Use of Knowledge Engineering for LearningGUIDON [Clancey 1982] exemplifies the use of a previously existing knowledge base or expert system as a teaching tool. GUIDON employs the rules from the MYCIN bacterial infection diagnoser to teach problem solving skills. More straightforward uses of expert systems as teaching systems can be exemplified by the work of King and McAulay [King & Mcaulay 1991] who used expert systems as a tool in teaching management students aspects of standard costing. Other researchers have investigated the utility of active participation in knowledge based system (KBS) development. Wideman and Owston [Wideman & Owston 1993] examined the gains in cognitive skills displayed by students involved in creating expert systems for weather prediction while Law and Ogburn [Law & Ogburn 1994] encouraged their students to build rule-bases describing commonsense knowledge about objects and motion. Trollip and Lippert [Trollip & Lippert 1987] used the development of a system for the design of Computer Aided Instruction screens to teach students what was involved in this kind of design. Other uses of expert systems (in Scottish schools) are reported in Conlon and Bowman [Conlon & Bowman 1995] and Conlon and Pain [Conlon & Pain 1996].Wideman and Owston offer qualified support for the "viability and efficacy of knowledge base creation as an instructional strategy for fostering cognitive development" (p. 194). Law and Ogburn point out (p. 511) that building expert systems "provided an opportunity to probe into a completely different aspect of the students'knowledge structures: the epistemic plane of operation represented by the elicited knowledge…" Trollip and Lippert (p. 46) describe the “process [a]s highly motivating” and conclude (p. 47): “The result for the student is not only a refinement of stored knowledge, but a refinement of cognitive skills and the conscious awareness of their extent and limitation”. On the other hand Clark [Clark 1990] declares that neither instruction with computers nor instruction in computing languages “makes any necessary psychological contribution to learning or to the transfer of what is learned during instruction” (p. 267). However, there is a prima facie case for assuming that the immersion of a learner in an environment which differentiates between different problem solving situations and indicates how these are linked to domain contents, will be of some use in overcoming the inadequacies in other approaches to the acquisition of meta-cognitive skills (especially skills involved in the selection of problem solving strategies).3.Constructivism and its ToolsIn the papers discussed above it is generally agreed that when students are given tools which allow them to construct their own versions of some real-life domain in interaction with other students and experts, their interest level and cognitive skills improve. These results are broadly compatible with the constructivist paradigm in educational practice. Constructivism can be seen as the view that meanings and knowledge are constructed by learners actively interpreting their experience [Nicaise & Barnes 1996; Jonassen et al. 1993a; Perkins 1991; Duffy & Jonassen 1991]. While social constructivists focus on this construction as a communal activity [see Prawat & Floden 1994], collaboration is part of constructivism generally. The implication for learning (as Prawat & Floden point out) is that teachers become facilitators for knowledge construction rather than knowledge transmitters: “This notion has led to calls for a dramatic shift in classroom focus away from the traditional transmission model of teaching toward one which is much more complex and interactive.” (p. 37) Many constructivists [e.g. Jonassen et al. 1993a; Knuth & Cunningham 1991] also espouse the idea that students should be expected to cope with “authentic” problems; an idea from research on situated cognition [Brown, Collins & Duguid 1989].The use of constructivist tools has been discussed by a variety of researchers [e.g., Duffy et al. 1991; Duffy & Jonassen 1991; Perkins 1991; Johassen et al. 1993a]. As a constructivist approach to knowledge gains ground over objectivism we can observe two main responses on the part of builders of computer systems for educational use: I will refer to them as the communication-centred and the development-centred.The communication-centred approach is exemplified by Chee's work on the system MIND BRIDGES [Chee 1996]. MIND BRIDGES is a system for "collaborative knowledge building" (p. 137) which allows students to collaborate on the building of a shared representation of their knowledge about some area. There are six "knowledge-building environments" and a range of "thematic spaces". Thematic spaces correspond to traditional disciplinary areas such as physics and literature while the environments (Overview, Explanation, Processes, Application, Conjecture, Location or Time) are included as a means of focusing the students communications on a particular aspect of a problem (such as conjectures about the habitability of Saturn in the astronomy thematic space, to use their example).The work of Johassen and his colleagues [Johassen et al. 1993a] can be described as development-centred. The applications they discuss “include expert systems as feedback facilitators, personal knowledge representation tools and cognitive study tools" (p. 93). As an example of the KBS as computer-based cognitive study tool, Jonassen and his colleagues outline the development of a rule base which models the decision-making used in deciding to authorise the Hiroshima bomb. They conclude (p. 93): “…it is clear that building expert systems requires learners to synthesize knowledge by making explicit their own reasoning”.Systems such as MIND BRIDGES represent a necessary but not sufficient response to the needs of the knowledge age. They provide the means of expressing knowledge models which, as Rowland [Rowland 1995] points out (p. 351), is necessary in the postmodern age. In addition, we need more complex tools for the creation of knowledge models. This need is met in the development-centred approach by coopting knowledge engineering ideas.4.Recent Developments in Knowledge EngineeringAs Conlon and Bowman [Conlon & Bowman 1995] point out (p. 129): “The current generation of educational shells is based on ideas that in wider KBS terms are out of date.” What are these developments? In contrast to earlier views which saw knowledge as a stuff to be extracted and poured into representation formalisms, recent research sees the process as one in which the engineer, in cooperation with the expert andusers, creates multiple models of the problem solving activities in a domain. The KADS knowledge engineering methodology [Schreiber et al. 1993; Tansley & Hayball 1993] is centred on the development of a series of models and is explicitly constructivist, at least in its recent forms [Breuker & Van de Velde 1994].Figure 1: Knowledge modellingUsing ideas from KADS, a knowledge model is constructed in the following manner. From an informal description of a domain and/or some problems associated with it, the modeller selects both a description of the problem to be solved (the problem type) using a library (with types such as Interpretation, Diagnosis and Planning), and a pre-defined template for the contents of the domain (the domain ontology). The domain ontology is composed of a vocabulary for a domain (e.g., for the medical domain, diseases, temperature levels and so on) and a schema [see Schreiber et al. 1993] or way of structuring domain contents in terms of relations (e.g., again in the medical domain, “Bacterial infection causes high temperature”). Using the problem type (and domain ontology), the modeller selects a means of carrying out the processing needed to solve the problem (the problem solving method or PSM) from a library of PSMs (such as Heuristic Classification, Systematic Refinement and Parametric Design). At the same time, using the domain ontology (and problem type), the modeller produces a detailed domain description; i.e. the facts, relations and heuristics which comprise the domain, couched in the vocabulary given by the ontology. Finally, the modeller produces a set of mappings between the problem solving method and the domain description. This is needed since the problem solving method uses its own terminology and these need to be associated with the appropriate parts of the domain description. For example, the problem solving term “symptom” might be associated with assertions of the form “There’s a problem with starting the car” in the car maintenance domain.5.Modelling Components IllustratedIt is possible to construct a knowledge model of the interpretation process used by archaeologists in identifying structures in air photos in terms of a sequence of primitive tasks (in this case, abstraction, matching and specialisation) which operate on a domain described in terms of expected structures (such as Romano-British military remains) and the features extracted from air photos (such as the shape, size, orientation and location of structures). In this case, the problem solution is largely a matter of mapping from the features to the expected structures.Problem Types: The problem type (or task) is given trivially by the description of the problem. In this instance the task is the identification (or classification) of the building structures contained in an air photo. In other archaeological problem solving situations the problem type might be planning (an excavation), assessment (of a theory), verification (of an hypothesis), design (of a series of excavation trenches) and so on. The distinction between problem types enables the learner to think clearly about the different types of activities carried out in a domain.Ontologies: The vocabulary which is relevant to the domain is an intersection of the set of terms for describing air photos (e.g., shadows, linear features, irregular features) and those for describing Romano-British building structures (e.g., domestic, military, commercial, fort, fortlet, camp). The domain schema here goes beyond the vocabulary to capture the possible relations between the domain concepts as given by the terms (e.g., fortlet is-smaller-than fort). The domain schema (and its instances) has much in common with the work on structural knowledge by Jonassen and his colleagues [Jonassen et al. 1993b]. While vocabularies provide the learner with a clear and consistent domain terminology which can be used in collaborative model construction, schemas provide a means of structuring the domain concepts and are part of the conceptual framework of any domain. Domain Descriptions: In accordance with the vocabulary and schema given above the domain description consists of a series of instantiated relations. The construction of these will involve the learner in a process of knowledge acquisition (in the engineering sense). In archaeology, the outcome of the process will be equivalent to the section of an archaeological report which includes full details of some aspect of the archaeological record. In addition, the engineering process assists in the acquisition of various relations between concepts which would not commonly figure in an archaeological report. This immersion of the learner in a domain should ensure that at least this static knowledge is retained.PSM: Given a view of air photo interpretation in which particular features (e.g., rectangular foundations) are evidence for particular archaeological structures when other evidence, such as the relation to other structures, is taken into account, the most likely problem solving family is that of Classification and, in particular, Heuristic Classification, since more than one of its reasoning components are clearly present—lines, areas and features are abstracted into higher level descriptions and descriptions mapped to structures. The basic structure for reasoning using Heuristic Classification is given below in [Fig. 2] which is based on that given in [Tansley & Hayball 1993]. See [Clancey 1985] for a full account. In [Fig. 2] ellipses represent tasks while rectangles represent domain roles (so-called because they indicate the sort of role to be played by domain items in problem solving).Figure 2: The PSM for Heuristic ClassificationWhether constructed or retrieved from libraries, PSMs are especially useful since they are not usually recorded or represented in normal disciplinary discourse. In addition, while PSMs are primarily models of problem solving skills, it is possible to model meta-cognitive processes such as PSM selection and validation since these are also problem solutions.Mappings: In this example, mappings need to be made between the PSM for Heuristic Classification shown in [Fig. 2] and the domain description. In this case there are two mappings needed for the domain roles Observables and Solutions. Here Observables map onto domain relations (such as, “structure11 has_shape square”) while Solutions maps onto relations such as “structure11 is_a auxiliary-fort”. We also need mappings between the three Heuristic Domain Roles, abstraction-knowledge, matching-knowledge and specialisation-knowledge, and domain-specific or generic heuristics. Mappings are useful for the learner in that they show how problem solving knowledge gets a purchase on domain knowledge and how different types of domain role play a variety of roles in problem solving.6.Learning as Knowledge ModellingAs Conlon and Bowman [Conlon & Bowman 1995] point out (following Clancey), “[t]he analogy between pupils and knowledge engineers can be taken too far” (p. 114). While there is a difference between knowledge engineering (where the goal is the production of models which can be used in the design and implementation of a fully working system) and education (where the goal is an increase in the ability of learners to differentiate between types of knowledge, to acquire new knowledge, to communicate about their knowledge and to applytheir skills in real life situations), nonetheless the following are the likely benefits to the learner in making use of knowledge modelling components:1. Knowledge models can be collaboratively developed (i.e. constructivism in practice) with opportunities forstudent role-playing;2. The model components and domain description vocabularies will provide students with a common language(which can be more or less precise depending on their educational level);3. Students interacting with knowledge level models will come to understand that there are many differenttypes of knowledge, that domain descriptions can be used for many different problems and that the same problem solving strategies can be used across a number of domains.4. These models are useful as a means of learning domain contents, problem solving skills, meta-cognitivestrategies, and domain discourse(s).7.Future workI have formulated some theoretical and practical questions which need to be settled before moving to a final version of any system based on the knowledge modelling approach:• How should relations (and other components) be represented (graphically, in natural language, in some formal notation) and how expressive should this notation be?• Does this approach help with the acquisition of problem solving skills and/or meta-cognitive skills such as problem solving strategy selection, or does it only teach modelling skills? That is, to what extent are meta-cognitive processes “only …relevant in the context of a particular learning process” as [Jonassen et al.1993a] claim (p. 92)?• To what extent does the basic modelling component set need to be extended for learner domain modelling?• Do different age groups and learners with different aims need different kinds of knowledge modelling system?• Are learners not able to benefit from theories of problem solving as [Conlon & Bowman 1995] imply (p.114)?In order to resolve at least some of these I intend to produce a new knowledge modelling tool which will combine constructivist ideas, knowledge engineering components and the results of research into computer supported argumentation (thus extending the rhetorical aspects of systems such as MINDBRIDGES). There will be three phases in the development of the tool. In the first, a knowledge modeller (akin to Jonassen’s personal knowledge representation tool) will be developed. This tool will include graphical and textual editors for models composed of the modelling components discussed above along with libraries of PSMs, problem types and domain ontologies, and an animation tool. The second phase will include learner collaboration while, in the third phase, a means of attaching the decision rationales (or arguments) of modellers will be implemented, thus completing our trivium by adding the full rhetorical dimension. This system will have much in common with that discussed in Conlon and Bowman [Conlon & Bowman 1995] and [Conlon & Pain 1996] where classification is selected as the most useful task (here a conflation of my problem type and PSM) to embody in a system for constructing learners’ knowledge bases. The tool I envisage differs in that the full range of modelling components, the entire set of problem types and a library of associated methods will be made available to students. This is partly for flexibility in modelling but also because it is more likely that problem solving and meta-cognitive skills will be acquired if a range of problem types and problem solving methods is encountered.8.ConclusionA tool built using recent ideas from knowledge engineering is not only largely consistent with constructivist ideas, it may provide the means for achieving some of the larger ambitions of constructivist researchers (such as the development of meta-cognitive skills). I say “largely consistent” since I am unsure to what extent the modelling framework provided by knowledge engineering is flexible enough for truly constructivist modelling. I suspect however that some structure (and therefore inflexibility) is needed if an environment is to act as a modelling environment. Given this, it is largely up to the learner to acquire the modelling and other cognitive skills necessary for the effective use of the new tool just as medieval students had to learn logic, rhetoric and grammar. The question of whether knowledge engineering provides the appropriate structure will be resolved when a system based on these ideas is built and tried out by learners.References[Breuker & Van de Velde 1994] Breuker, J. A. & Van de Velde, W. (1994). eds. The CommonKADS Library for Expertise Modelling, Amsterdam: IOS Press.[Brown, Collins & Duguid 1989] Brown, J.S., Collins, A. & Duguid, P. (1989). Situated Cognition and the Culture of Learning. Educational Researcher, 18(1), 32–42.[Chee 1996] Chee, Y.S. (1996). MIND BRIDGES: A Distributed, Multimedia Learning Environment for Collaborative Knowledge Building. International Journal of Educational Telecommunications, 2(2/3), 137-153.[Clancey 1982] Clancey, W.J. (1982). Tutoring rules for guiding a case method dialogue. In Sleeman, D. & Brown, J.S. eds. Intelligent Tutoring Systems. London: Academic Press. 201–225.[Clancey 1985] Clancey, W.J. (1985). Heuristic Classification. Artificial Intelligence, 27, 289–350.[Clark 1990] Clark, R.E. (1990). Facilitating Domain-General Problem Solving: Computers, Cognitive Processes and Instruction. In de Corte, E., Linn, M.C., Mandl, H. & Verschaffel, L., eds. Computer-Based Learning Environments and Problem Solving. Berlin: Springer-Verlag. 265–285.[Conlon & Bowman 1995] Conlon, T. & Bowman, N. (1995). Expert Systems, Shells, and Schools: Present Practice, Future Prospects. Instructional Science, 23, 111–131.[Conlon & Pain 1996] Conlon, T. & Pain, H. (1996). Persistent Collaboration: A Methodology for Applied AIED. Journal of Artificial Intelligence in Education, 7(3/4), 219–252.[Duffy & Jonassen 1991] Duffy, T.M. & Jonassen, D.H. (1991). Constructivism: New Implications for Instructional Technology? Educational Technology, 31(5), 7–12.[Duffy et al. 1991] Duffy, T.M., Lowyck, J., Jonassen, D.H. & Welsh, T.M. (1991). Designing Environments for Constructive Learning. Berlin: Springer-Verlag.[Eco 1987] Eco, U. (1987). Travels in Hyperreality: Essays. London: Picador-Pan.[Jonassen et al. 1993a] Jonassen, D.H., Wilson, B.G., Wang, S. & Grabinger, R.S. (1993). Constructivist Uses of Expert Systems to Support Learning. Journal of Computer-Based Instruction, 20(3), 86-94.[Jonassen et al. 1993b] Jonassen, D.H., Beissner, K. and Yacci, M. (1993). Structural Knowledge: Techniques for Representing, Conveying, and Acquiring Structural Knowledge. Hillsdale, NJ: Lawrence Erlbaum.[King & McAulay 1991] King, M. and McAulay, L. (1991). Experiments with expert systems in management education. Journal of Information Technology, 6, 34–38.[Knuth & Cunningham 1991] Knuth, R.A. and Cunningham, D.J. (1991). Tools for Constructivism. In [Duffy et al. 1991]. 163–188.[Law & Ogburn 1994] Law, N. and Ogburn, J. (1994). Students as Expert System Developers: A Means of Eliciting and Understanding Commonsense Reasoning. Journal of Research on Computing in Education, 26(4), 497-513. [Nicaise & Barnes 1996] Nicaise, M. & Barnes, D. (1996). The union of technology, constructivism, and teacher education. Journal of Teacher Education, 47, 205–212.[Perkins 1991] Perkins, D.N. (1991). Technology meets Constructivism: Do they make a marriage? Educational Technology, 31(5), 18–23.[Prawat & Floden 1994] Prawat, R.S. & Floden, R.E. (1994). Philosophical Perspectives on Constructivist Views of Learning. Educational Psychology, 29(1), 37–48.[Rowland 1995] Rowland, R.C. (1995). In Defense of Rational Argument: A Pragmatic Justification of Argumentation Theory and Response to the Postmodern Critique. Philosophy and Rhetoric, 28(4), 350–364.[Schreiber et al. 1993] Schreiber, G., Wielinga, B. & Breuker, J, eds. (1993). KADS: A Principled Approach to Knowledge-Based System Development, London: Academic Press.[Tansley & Hayball 1993] Tansley, D.S.W. & Hayball, C.C. (1993). Knowledge-Based Systems Analysis and Design: A KADS Developer's Handbook. Hemel Hempstead: Prentice Hall.[Trollip & Lippert 1987] Trollip, S.R. & Lippert, R.C. (1987). Constructing Knowledge Bases: A Promising Instructional Tool. Journal of Computer-Based Instruction, 14(2), 44–48.[Wideman & Owston 1993] Wideman, H.H. & Owston, R.D. (1993). Knowledge Base Construction as a Pedagogical Activity. J. Educational Computing Research, 9(2), 165-196.AcknowledgementsMany thanks to Enrico Motta and Stuart Watt from the Knowledge Media Institute for commenting on earlier versions of this paper. I have incorporated many of their suggestions.。

Logics for Knowledge RepresentationBernhard NebelAlbert-Ludwigs-Universit¨a t Freiburg,Germany1IntroductionKnowledge representation and reasoning plays a central role in Artificial Intelli-gence.Research in Artificial Intelligence(henceforth AI)started off by trying to identify the general mechanisms responsible for intelligent behavior.However,it quickly became obvious that general and powerful methods are not enough to get the desired result,namely,intelligent behavior.Almost all tasks a human can per-form which are considered to require intelligence are also based on a huge amount of knowledge.For instance,understanding and producing natural language heavily relies on knowledge about the language,about the structure of the world,about social relationships etc.One way to address the problem of representing knowledge and reasoning about it is to use some form of logic.While this seems to be a natural choice,it took a while before this“logical point of view”became the prevalent approach in the area of knowledge representation.Below,we will give a brief sketch of how thefield of knowledge representation evolved and what kind logical methods have been used. Int.Encyc.Social and Behavioral Sciences27February2001In particular,we will argue that the important point about using formal logic is the logical method.2Logic-Based Knowledge Representation:A Historical AccountMcCarthy(1968)stated very early on that mathematical,formal logic appears to be a promising tool for achieving human-level intelligence on computers.In fact, this is still McCarthy’s(2000)vision,which he shares with many researchers in AI.However,in the early days of AI,there were also a number of researchers with a completely different opinion.Minsky(1975),for example,argued that knowl-edge representation formalisms should beflexible and informal.Moreover,he ar-gued that the logical notions of correctness and completeness are inappropriate in a knowledge representation context.While in those days heated arguments of the suitability of logic were exchanged,by the end of the eighties,the logical perspective seem to have gained the upper hand (Brachman1990).During the nineties almost all research in the area of knowledge representation and reasoning was based on formal,logical methods as demonstrated by the papers published in the bi-annual international conference on Principles of Knowledge Representation and Reasoning,which started in1989.It should be noted,however,that two perspectives on logic are possible.Thefirst perspective,taken by McCarthy(1968),is that logic should be used to represent knowledge.That is,we use logic as the representational and reasoning tool insidethe computer.Newell(1982)on the other hand proposed in his seminal paper on the knowledge level to use logic as a formal tool to analyze knowledge.Of course, these two views are not incompatible.Furthermore,once we accept that formal logic should be used as a tool for analyzing knowledge,it is a natural consequence to use logic for representing knowledge and for reasoning about it as well.3Knowledge Representation Formalisms and Their SemanticsSaying that logic is used as the main formal tool does not say which kind of logic is used.In fact,a large variety of logics(Gabbay,Hogger and Robinson1995)have been employed or developed in order to solve knowledge representation and rea-soning problems.Often,one started with a vaguely specified problem,developed some kind knowledge representation formalism without a formal semantics,and only later started to provide a formal ing this semantics,one could then analyze the complexity of the reasoning problems and develop sound and complete reasoning algorithms.I will call this the logical method,which proved to be very fruitful in the past and has a lot of potential for the future.3.1Description LogicsOne good example for the evolution of knowledge representation formalisms is the development of description logics,which have their roots in so-called struc-tured inheritance networks formalisms such as KL-ONE(Brachman1979).These networks were originally developed in order represent word meanings.A conceptnode connects to other concept nodes using roles.Moreover,the roles could be structured as well.These networks permits for,e.g.,the definition of the concept of a bachelor.Later on,these structured inheritance networks were formalized as so-called con-cept languages,terminological logics,or description logics.Concepts were inter-preted as unary predicates,roles as binary relations,and the connections between nodes as so-called value restrictions.This leads for most such description logics to a particular fragment offirst-order predicate logic,namely,the fragment.In this fragment only two different variable symbols are used.As it turns out,this is a decidable fragment offirst-order logic.However,some of the more involved description logics go beyond.They con-tain,e.g.,relational composition or transitive closure.As it turns out,such descrip-tion logics can be understood as variants of multi-modal logics(Schild1991),and decidability and complexity results from these multi-modal logics carry over to the description logics.Furthermore,description logics are very close to feature log-ics as they are used in unification-based grammars.In fact,description logics and feature logics can be viewed as members of the same family of representation for-malisms(Nebel and Smolka1990).All these insights,i.e.,determination of decidability and complexity as well as the design of decision algorithms(e.g.Donini,Lenzerini,Nardi and Nutt1991),are based on the rigorous formalization of the initial ideas.In particular,it is not justone logic that it is used to derive these results,but it is the logical method that led to the success.One starts with a specification of how expressions of the language or formalism have to be interpreted in formal terms.Based on that one can specify when a set of formulae logically implies a formula.Then one can start tofind similar formalisms(e.g.modal logics)and prove equivalences and/or one can specify a method to derive logically entailed sentences and prove them to be correct and complete.3.2Nonmonotonic LogicsAnother interesting area where the logical method has been applied is the devel-opment of the so-called non-monotonic logics.These are based on the intuition that sometimes a logical consequence should be retracted if new evidence becomes known.For example,we may assume that our car will not be moved by somebody else after we have parked it.However,if new information becomes known,such as the fact that the car is not at the place where we have parked it,we are ready to drop the assumption that our car has not been moved.This general reasoning pattern was used quite regularly in early AI systems,but it took a while before it was analyzed from a logical point of view.In1980,a special issue of the Artificial Intelligence journal appeared,presenting different ap-proaches to non-monotonic reasoning,in particular Reiter’s(1980)default logic and McCarthy’s(1980)circumscription approach.A disappointing fact about nonmonotonic logics appears to be that it is very difficult to formalize a domain such that one gets the intended conclusions.In particular,in the area of reasoning about actions,McDermott(1987)has demonstrated that the straightforward formalization of an easy temporal projection problem(the“Yale shooting problem”)does not lead to the desired consequences.However,it is pos-sible to get around this problem.Once all underlying assumptions are spelled out, this and other problem can be solved(Sandewall1994).It took more than a decade before people started to analyze the computational com-plexity(of the propositional versions)of these logics.As it turned out,these log-ics are usually somewhat more difficult than ordinary propositional logic(Gottlob 1992).This,however,seems tolerable since we get much more conclusions than in standard propositional logic.Right at the same time,the tight connection between nonmonotonic logic and belief revision(G¨a rdenfors1988)was noticed.Belief revision–modeling the evolution of beliefs over time–is just one way to describe how the set of nonmonotonic consequences evolve over time,which leads to a very tight connection on the formal level for these two forms of nonmonotonicity(Nebel1991).Again,all these results and insights are mainly based on the logical method to knowledge representation.4OutlookThe above description of the use of logics for knowledge representation is nec-essarily incomplete.For instance,we left out the area of qualitative temporal and spatial reasoning completely.Nevertheless,one should have got an idea of how log-ics are used in the area of knowledge representation.As mentioned,it is the idea of providing knowledge representation formalisms with formal(logical)semantics that enables us to communicate their meaning,to analyze their formal properties, to determine their computational complexity,and to devise reasoning algorithms.While the research area of knowledge representation is dominated by the logical approach,this does not mean that all approaches to knowledge representation must be based on logic.Probabilistic(Pearl1988)and decision theoretic approaches, for instance,have become very popular lately.Nowadays a number of approaches aim at unifying decision theoretic and logical accounts by introducing a qualita-tive version of decision theoretic concepts(Benferhat,Dubois,Fargier,Prade and Sabbadin2000).Other approaches(Boutilier,Reiter,Soutchanski and Thrun2000) aim at tightly integrating decision theoretic concepts such as Markov decision pro-cesses with logical approaches,for instance.Although this is not pure logic,the two latter approaches demonstrate the generality of the logical method:specify the formal meaning and analyze!BibliographyAllen,J. A.,Fikes,R.and Sandewall, E.(eds):1991,Principles of Knowledge Representation and Reasoning:Proceedings of the2nd International Conference(KR-91),Morgan Kaufmann,Cambridge,MA.Benferhat,S.,Dubois,D.,Fargier,H.,Prade,H.and Sabbadin,R.:2000,Decision, nonmonotonic reasoning and possibilistic logic,in Minker(2000),pp.333–360. Boutilier,C.,Reiter,R.,Soutchanski,M.and Thrun,S.:2000,Decision-theoretic,high-level agent programming in the situation calculus,Proceedings of the17th National Conference of the American Association for Artificial Intelligence(AAAI-2000),MIT Press,Austin,TX.Brachman,R.J.:1979,On the epistemological status of semantic networks,in N.V.Findler (ed.),Associative Networks:Representation and Use of Knowledge by Computers, Academic Press,New York,NY,pp.3–50.Brachman,R.J.:1990,The future of knowledge representation,Proceedings of the8th National Conference of the American Association for Artificial Intelligence(AAAI-90),MIT Press,Boston,MA,pp.1082–1092.Donini,F.M.,Lenzerini,M.,Nardi,D.and Nutt,W.:1991,The complexity of concept languages,in Allen,Fikes and Sandewall(1991),pp.151–162.Gabbay,D.M.,Hogger,C.J.and Robinson,J.A.(eds):1995,Handbook of Logic in Artificial Intelligence and Logic Programming–Vol.1–5,Oxford University Press, Oxford,UK.G¨a rdenfors,P.:1988,Knowledge in Flux—Modeling the Dynamics of Epistemic States, MIT Press,Cambridge,MA.Gottlob,G.:1992,Complexity results for nonmonotonic logics,Journal for Logic and Computation2(3),397–425.McCarthy,J.:1968,Programs with common sense,in M.Minsky(ed.),Semantic Information Processing,MIT Press,Cambridge,MA,pp.403–418.McCarthy,J.:1980,Circumscription—a form of non-monotonic reasoning,Artificial Intelligence13(1–2),27–39.McCarthy,J.:2000,Concepts of logical AI,in Minker(2000),pp.37–58. McDermott,D.V.:1987,A critique of pure reason,Computational Intelligence3(3),151–160.Minker,J.(ed.):2000,Logic-Based Artificial Intelligence,Kluwer,Dordrecht,Holland. Minsky,M.:1975,A framework for representing knowledge,in P.Winston(ed.),The Psychology of Computer Vision,McGraw-Hill,New York,NY,pp.211–277. Nebel,B.:1991,Belief revision and default reasoning:Syntax-based approaches,in Allen et al.(1991),pp.417–428.Nebel,B.and Smolka,G.:1990,Representation and reasoning with attributive descriptions, in K.-H.Bl¨a sius,U.Hedtst¨u ck and C.-R.Rollinger(eds),Sorts and Types in Artificial Intelligence,V ol.418of Lecture Notes in Artificial Intelligence,Springer-Verlag, Berlin,Heidelberg,New York,pp.112–139.Newell,A.:1982,The knowledge level,Artificial Intelligence18(1),87–127.Pearl,J.:1988,Probabilistic Reasoning in Intelligent Systems:Networks of Plausible Inference,Morgan Kaufmann,San Francisco,CA.Reiter,R.:1980,A logic for default reasoning,Artificial Intelligence13(1),81–132. Sandewall,E.:1994,Features and Fluents,Oxford University Press,Oxford,UK. Schild,K.:1991,A correspondence theory for terminological logics:Preliminary report, Proceedings of the12th International Joint Conference on Artificial Intelligence (IJCAI-91),Morgan Kaufmann,Sydney,Australia,pp.466–471.。

Ontology-based Reasoning about Lexical ResourcesJan Scheffczyk,Collin F.Baker,Srini NarayananInternational Computer Science Institute1947Center St.,Suite600,Berkeley,CA,94704jan,collinb,snarayan@AbstractReasoning about natural language most prominently requires combining semantically rich lexical resources with world knowledge, provided by ontologies.Therefore,we are building bindings from FrameNet–a lexical resource for English–to various ontologies depending on the application at hand.In this paper we show thefirst step toward such bindings:We translate FrameNet to the Web Ontology Language OWL DL.That way,FrameNet and its annotations become available to Description Logic reasoners and other OWL tools.In addition,FrameNet annotations can provide a high-quality lexicalization of the linked ontologies.1.IntroductionCombining large lexical resources with world knowledge, via ontologies,is a crucial step for reasoning over natu-ral language,particularly for the Semantic Web.Concrete applications include semantic parsing,text summarization, translation,and question answering.For example,ques-tions like“Could Y have murdered X?”may require sev-eral inference steps based on semantic facts that simple lexicons do not include.Moreover,they require so-called open-world semantics offered by state-of-the art Descrip-tion Logic(DL)reasoners,e.g.,FaCT(Horrocks,1998) or Racer(Wessel and M¨o ller,2005).The FrameNet lex-icon(Ruppenhofer et al.,2005)has a uniquely rich level of semantic detail;thus,we are building bindings from FrameNet to multiple ontologies that will vary depending on the application.That way,we enable reasoners to make inferences over natural-language text.In this paper,we report on thefirst step toward this goal:we have automatically translated a crucial portion of FrameNet to OWL DL and we show how state-of-the-art DL reasoners can make inferences over FrameNet-annotated sentences. Thus,annotated text becomes available to the Semantic Web and FrameNet itself can be linked to other ontologies. This work gives a clear motivation for the design of our pro-posed ontology bindings and defines the baseline for mea-suring their benefits.This paper proceeds as follows:In Sect.2.we briefly intro-duce FrameNet–a lexical resource for English.We present our design decisions for linking FrameNet to ontologies in Sect.3.Sect.4.includes the heart of this paper:A formal-ization of FrameNet and FrameNet-annotated sentences in OWL DL.In Sect.5.we show how our OWL DL represen-tation can be used by the DL reasoner RacerPro in order to implement tasks of a question answering system,based on reasoning.We evaluate our approach in Sect.6.Sect.7. concludes and sketches directions for future research.2.The FrameNet Lexicon FrameNet is a lexical resource for English,based on frame semantics(Fillmore,1976;Fillmore et al.,2003; Narayanan et al.,2003).A semantic frame(hereafter sim-ply frame)represents a set of concepts associated with an event or a state,ranging from simple(Arriving,Placing)to complex(Revenge,Criminalaffect(which in turn inherits from the frames Transitiveact).In addition,Attack uses the frame Hos-tileact frame.3.Linking FrameNet to Ontologies forReasoningNLP applications using FrameNet require knowledge about the possiblefillers for FEs.For example,a semantic frame parser needs to know whether a certain chunk of text(or a named entity)might be a properfiller for an FE–so it will check whether thefiller type of the FE is compatible with the type of the named entity.Therefore,we want to provide constraints onfillers of FEs,so-called semantic types(STs). Currently,FrameNet itself has defined about40STs that are ordered by a subtype hierarchy.For example,the Assailant FE and the Victim FE in the Attack frame both have the ST Sentient,which in turn is a subtype of Animatething,Physical entity. It is obvious that FrameNet STs are somewhat similar to the concepts(often called classes)defined in ontologies like SUMO(Niles and Pease,2001)or Cyc(Lenat,1995). Compared to ontology classes,however,FrameNet STs are much more shallow,have fewer relations between them(we only have subtyping and no other relations),and are notFigure1:Abridged example frame Attack and some connected frames.context specific.Naturally,in a lexicographic project like FrameNet STs play a minor role only.Therefore,we want to employ the STs from existing large ontologies such as SUMO or Cyc;in this way we will gain a number of advantages almost for free:AI applications can use the knowledge provided by the target ontology.We can provide different STs suitable for particular applications by bindings to different ontologies.We can use ontologies in order to query and analyze FrameNet data.For example,we can measure the se-mantic distance between frames based on different tar-get ontologies or we can check consistency and com-pleteness of FrameNet w.r.t.some target ontology.The target ontologies would benefit from FrameNet, supplementing their ontological knowledge with a proper lexicon and annotated example sentences. Compared to other lexicon-ontology bindings(Niles and Pease,2003;Burns and Davis,1999),our bindings offer a range of advantages due to specific FrameNet character-istics:FrameNet models semantic and syntactic valences plus the predicate-argument structure.FrameNet includes many high-quality annotations,providing training data for machine learning.In contrast to WordNet synset annota-tions,our annotations include role labelling.Frame seman-tics naturally provides cross-linguistic abstraction plus nor-malization of paraphrases and support for null instantiation (NI).Notice that a detour via WordNet would introduce ad-ditional noise through LU lookup(Burchardt et al.,2005). In addition,WordNet synset relations are not necessarily compatible with FrameNet relations.The bindings from FrameNet to ontologies should be de-scribed in the native language of the target ontologies,i.e., KIF(for bindings to SUMO),CycL(for bindings to Cyc), or OWL(for bindings to OWL ontologies).This allows the use of standard tools like reasoners directly,without any intermediate steps.Also,arbitrary class expressions can be used and ad-hoc classes can be defined if no exact corre-sponding class could be found in the target ontology.We expect this to be very likely because FrameNet is a lexico-graphic project as opposed to ontologies,which are usually driven by a knowledge-based approach.Finally,the bind-ing should be as specific as possible for the application at hand.For example,in a military context we would like to bind FEs to classes in an ontology about WMD or terror-ism instead of using a binding to SUMO itself,which only provides upper level classes.2The vital precondition for any such bindings is,however, to have FrameNet available in an appropriate ontology language(e.g.,KIF,CycL,or OWL).A representation of FrameNet in an ontology language bears the additional ad-vantages of formalizing certain properties of frames and FEs,and enabling us to use standard tools to view,query, and reason about FrameNet data.For querying,one could, e.g.,use the ontology query language SPARQL.Next,we describe a formalization of a portion of FrameNet in OWL DL,which easily generalizes to more expressive ontology languages like KIF or CycL.4.Formalizing FrameNet in OWL DL Our major design decisions for representing FrameNet as an ontology are:1.to represent frames,FEs,and STs formally as classes,2.to model relations between frames and FEs via exis-tential property restrictions on these classes,and3.to represent frame and FE realizations in FrameNet-annotated texts as instances of the appropriate frame and FE classes,respectively.Building on(Narayanan et al.,2003),we have chosen OWL DL as representation language mainly because better tools are available for it(particularly for reasoning)than for OWL Full or other similarly expressive languages.Our representation differs from many WordNet OWL represen-tations,which represent synsets as instances and hence can-not use class expressions for ontology bindings.3Instead, WordNet bindings to SUMO employ a proprietary mecha-nism,4which cannot be used“out of the box”by ontology tools like reasoners.In order to keep the size of our ontology manageable, we have chosen to split it into the FrameNet Ontology and Annotation Ontologies.The FrameNet Ontology in-cludes FrameNet data like frames,FEs,and relations be-tween them.Annotation Ontologies represent FrameNet-annotated sentences and include parts of the FrameNet On-tology that are necessary.4.1.The FrameNet OntologyFig.2shows a simplified excerpt of the FrameNet On-tology.The subclasses of the Syntax class are used for annotations and are connected to frames and FEs via the evokes andfillerOf relations,respectively.Frames and FEs are connected via binary relations,e.g.,the usesF prop-erty or the hasFE property,which connects a frame to its FEs.Consider our example frame Attack,which inher-its from the frame Intentionallyencounter.We model frame and FE inheritance via subclassing and other frame and FE relations via existen-tial property restrictions(owl:someValuesFrom).Thus,the class Attack is a subclass of Intentionallyencounter connected via the usesF property.The FEs of Attack are connected via an ex-istential restriction on the hasFE property.FE relations are modeled similarly to frame relations.Recall that class restrictions are inherited.Therefore,the class Attack inherits the restrictions hasFE Patient and hasFE Agent from the class Intentionallyaffect has exactly one instance of type Patient con-nected via the hasFE property:Intentionallyaffect hasFE or Intentionally5see /2001/sw/BestPractices/OEP/QCR/6Even now,the FrameNet Ontology reaches a critical size of 100,000triples.class Sentient.We intend to use this mechanism for link-ing FrameNet to other ontologies also.So we can use ar-bitrary OWL DL class expressions for our bindings and at the same time achieve a homogeneous formal representa-tion that OWL tools can make use of.One could use the FrameNet Ontology for querying and reasoning over FrameNet itself.For reasoning over natu-ral language text,however,we mustfind a way to incorpo-rate this text into the FrameNet Ontology.We do this by means of Annotation Ontologies,which we generate from FrameNet-annotated text.4.2.Annotation OntologiesFrameNet-annotated text provides textual realizations of frames and FEs,i.e.,the frames and FEs cover the se-mantics of the annotated sentences.In ontological terms, FrameNet-annotated text constitutes instances of the appro-priate frame and FE classes,respectively.From an anno-tated sentence we generate an Annotation Ontology,which includes parts of the FrameNet Ontology and fulfills all its class restrictions.In other words,the FrameNet Ontology provides a formal specification for Annotation Ontologies. Consider an example sentence,which we derived from an evaluation exercise within the AQUINAS project called “KB Eval;”where sentences for analysis were contributed by various members of the consortium.S48Kuwaiti jetfighters managed to escape the Iraqi invasion.7This sentence has three annotation sets:1.The target word invasion evokes the Attack frame,where Iraqifills the Assailant FE.The Victim FE has nofiller,i.e.,it is null instantiated(NI).2.The target word escape evokes the Avoiding frame,with FEfillers48Kuwaiti jetfighters Agent,the Iraqi invasion Undesirableaction frame,with FEfillers48Kuwaiti jetfightersProtagonist,to escape the Iraqi invasion Goal. From this annotated sentence wefirst create a syntactic de-pendency graph and generate the appropriate frame and FE instances as shown in Fig.3A Span represents a chunk of text that can evoke a frame or provide afiller for an FE. We derive Spans,syntactic subsumption,and the relations to frames and FEs based on the annotations.For example, invasion evokes the Attack frame.Thus we(1)generate a Span that represents the text invasion and place it properly into the Span dependency graph,(2)generate the frame in-stance Attack S(with type Attack),and(3)connect the Span to Attack S via the evokes property.We proceed similarly with the FEfiller Iraqi Agent.Here we generate the FE instance Agent S,connect it to its frame instance Attack S via the hasFE property,and connect the Span representing Iraqi to Agent S via thefillerOf property.Finally,we iden-tify FEs that are evoked by the same Span via owl:sameAs.Figure2:Part of the FrameNet Ontology for the Attack frame and some connected frames.Figure3:Annotation Ontology for:48Kuwaiti jetfighters managed to escape the Iraqi invasion.(Step1)We can do this purely based on syntactic evidence.For ex-ample,the FE instances Protagonist S and Agent S are iden-tified because the are bothfilled by the Span representing the text48Kuwaiti jetfighters.This significantly aids rea-soning about FrameNet-annotated text.8The second step in generating an Annotation Ontology is to satisfy the class restrictions of the FrameNet ontology, i.e.,to generate appropriate instances and to connect them properly.Thus,for a frame instance of type we1.travel along each existential class restriction on a prop-erty to a class(),2.generate an instance of type,3.connect the instances and via the property,and4.proceed with instance.Fig.4illustrates this algorithm for our example frame instance Attack.We generate the frame instance Hos-tile1S and Sideencounter S via usesF.Sim-ilarly,we connect Assailant S to Side2S via usesFE.In addition,we identify the connected FE instances via owl:sameAs,which expresses the seman-tics of FE mappings:The Victim in an Attack is the Side encounter,i.e.,theirfillers are the same.In addition to the class restrictions,we also travel along the inheritance hierarchy,which could be useful,e.g.,for paraphrasing.Therefore,we generate the instance Inten-tionally8Alternatively,we could formalize a SWRL rule fillerOffillerOf owl:sameAs.We do not do so because not all reasoners provide a SWRL implementa-tion.9see succeeds,then the Spans bound to these FEs contain the an-swer,otherwise the question cannot be answered from the text.Consider three example questions.Q1How many Kuwaiti jetfighters escaped the Iraqi inva-sion?Q2How many Kuwaiti jetfighters escaped?Q3Did Iraq clash with Kuwait?Q4Was there a conflict between Iraq and Kuwait? Partial Annotation Ontologies for these questions are illus-trated in Fig.5.Given the Annotation Ontology of the question,we let Rac-erPro perform the following queries,which can be formal-ized in nRQL.10In the following we will use question Q1 as an example of how the algorithm works.1.For the question get the evoked frames instances,theirFEs,and Spans.Avoiding Q1Undesirables.Q1Undesirables.Undesirables.S the Iraqi invasionAgent S48Kuwaiti jetfightersAssailant S IraqiVictim S NIFigure4:Connecting the Attack instance(Step2of Annotation Ontologygeneration) Figure5:Abridged Annotation Ontologies for example questionsSince RacerPro is a reasoner(and no NLP tool), checking the compatibility of Spans is limited to checking syntactic equality.Therefore,the Span48 Kuwaiti jetfighters does not match the Span How many Kuwaiti jetfighters.We can,however,easily de-termine the Spans that are supposed to be compatible in order to yield an answer.Then Span compatibility can be determined by other NLP tools such as question type recognizers.Question Q2is simpler than Q1because we are asking for only one frame in which one FE is null instantiated.In this case our approach only using a reasoning engine yields the final answer:Undesirableencounter.Our ap-proach proceeds as follows:1.Get evoked frames instances,FEs,and Spans:Hostile1Q3IraqSideencounter Q3Hostile1Q3Side2Q3Sideencounter Hostile1Side2Side1S IraqiSide1S isthe same as Assailant S and Side1and Side1and Side1and Side11see /services/named entity recognizers.Also,we plan to evaluate the utility of DL reasoners in a fullyfledged question answer-ing system.Finally,we will translate FrameNet to other ontology languages such as KIF or CycL,in order to link FrameNet to SUMO or Cyc ontologies.AcknowledgementsThefirst author enjoys funding from the German Aca-demic Exchange Service(DAAD).The FrameNet project is funded by the AQUINAS project of the AQUAINT pro-gram.8.ReferencesA.Burchardt,K.Erk,and A.Frank.2005.A WordNet detour to FrameNet.In Proceedings of the GLDV2005 Workshop GermaNet II,Bonn.K.J.Burns and A.R.Davis.1999.Building and maintain-ing a semantically adequate lexicon using cyc.In Eve-lyne Viegas,editor,Breadth and Depth of Semantic Lex-icons.Kluwer.K.Erk and S.Pad´o.2005.Analysing models for semantic role assignment using confusability.In Proceedings of HLT/EMNLP-05,Vancouver,Canada.K.Erk and S.Pad´o.2006.Shalmaneser–a toolchain for shallow semantic parsing.In Proceedings of LREC-06, Genova,Italy.to appear.C.J.Fillmore,C.R.Johnson,and M.R.L.Petruck.2003. Background to FrameNet.International Journal of Lex-icography,16(3):235–250.C.J.Fillmore.1976.Frame semantics and the nature of language.Annals of the New York Academy of Sciences, (280):20–32.I.Horrocks.1998.The FaCT system.In H.de Swart, editor,Automated Reasoning with Analytic Tableaux and Related Methods:International Conference Tableaux’98,number1397in Lecture Notes in Artificial Intelligence,pages307–312.Springer-Verlag,May. D.B.Lenat.1995.Cyc:a large-scale investment in knowl-edge mun.ACM,38(11):33–38.K.Litowski.2004.Senseval-3task:Automatic labeling of semantic roles.In Senseval-3:Third International Work-shop on the Evaluation of Systems for the Semantic Anal-ysis of Text,pages9–12.Association for Computational Linguistics.S.Narayanan and S.McIlraith.2003.Analysis and simu-lation of Web works,42(5):675–693.S.Narayanan,C.F.Baker,C.J.Fillmore,and M.R.L. Petruck.2003.FrameNet meets the semantic web:Lexi-cal semantics for the web.In The Semantic Web—ISWC 2003,pages771–787.Springer-Verlag,Berlin.S.Narayanan.1999.Moving right along:A computational model of metaphoric reasoning about events.In Pro-ceedings of the/National Conference on Artificial Intel-ligence(AAAI’99),pages121–128.AAAI Press.I.Niles and A.Pease.2001.Towards a standard upper on-tology.In Proceedings of the2nd International Confer-ence on Formal Ontology in Information Systems(FOIS-2001),Ogunquit,Maine.I.Niles and A.Pease.2003.Linking lexicons and ontolo-gies:Mapping wordnet to the suggested upper merged ontology.In Proceedings of the2003International Conference on Information and Knowledge Engineering (IKE03).J.Ruppenhofer,M.Ellsworth,M.R.Petruck, and C.R.Johnson,2005.FrameNet: Theory and Practice.ICSI Berkeley. /framenet/book/book.html.J.Scheffczyk and M.Ellsworth.2006.Improving the qual-ity of FrameNet.In Proc.of the Wkshp.on Quality assurance and quality measurement for language and speech resources,Genoa,Italy.to appear.M.Wessel and R.M¨o ller.2005.A high performance se-mantic web query answering engine.In Proc.Interna-tional Workshop on Description Logics.。

Ontologies and the Configuration of Problem-Solving MethodsRudi Studer1,Henrik Eriksson2,John Gennari3,Samson Tu3, Dieter Fensel1,4,and Mark Musen31Institute AIFB, University of Karlsruhe, D-76128 Karlsruhee-mail: studer@aifb.uni-karlsruhe.de2Department of Computer and Information Science, Linköping University, S-58183 Linköpinge-mail: her@ida.liu.se3Section on Medical Informatics, Knowledge Systems Laboratory, Stanford University School of Medicine, Stanford, CA 94305-5479, USAe-mail: {gennari,tu,musen}@4Department SWI, University of Amsterdam, NL-1018 WB Amsterdame-mail: dieter@swi.psy.uva.nlAbstractProblem-solving methods model the problem-solving behavior of knowledge-based systems. The PROTÉGÉ-II framework includes a library of problem-solving methods that can be viewed as reusable components. For developers to use these components as building blocks in the construction of methods for new tasks, they must configure the components to fit with each other and with the needs of the new task. As part of this configuration process, developers must relate the ontologies of the generic methods to the ontologies associated with other methods and submethods. We present a model of method configuration that incorporates the use of several ontologies in multiple levels of methods and submethods, and we illustrate the approach by providing examples of the configuration of the board-game method.1. IntroductionProblem-solving methods for knowledge-based systems capture the problem-solving behavior required to performing the system's task (McDermott, 1988). Because certain tasks are common (e.g., planning and configuration), and are approachable by the same problem-solving behavior, developers can reuse problem-solving methods in several applications ((Chandrasekaran and Johnson, 1993), (Breuker and Van de Velde, 1994)). Thus, a library of reusable methods would allow the developer to create new systems by selecting, adapting and configuring such methods. Moreover, development tools, such as PROTÉGÉ-II (Puerta et al., 1992), can support the developer in the reuse of methods.Problem-solving methods are abstract descriptions of problem-solving behavior. The development of problem solvers from reusable components is analogous to the general approach of software reuse. In knowledge engineering as well as software engineering, developers often duplicate work on similar software components, which are used in different applications. The reuse of software components across several applications is a potentially useful technique that promises to improve the software-development process (Krueger, 1992). Similarly, the reuse of problem-solving methods can improve the quality, reliability, and maintainability of the software (e.g., by the reuse of quality-proven components). Of course, software reuse is only financially beneficial in the end if the indexing and configuration overhead is less than the effort that is needed to create the required component several times from scratch.Although software reuse is an appealing approach theoretically, there are serious practical problems associated with reuse. Two of the most important impediments to software reuse are (1) the problem of finding reusable components (e.g., locating appropriate components in a library), and (2) the problem of adapting reusable components to their task and to their environment. The firstproblem is sometimes called the indexing problem, and the second problem is sometimes called the configuration problem. These problems are also present in the context of reusable problem-solving methods. In the remainder of this paper we shall focus on the configuration problem.Method configuration is a difficult task, because the output of one method may not correspond to the input of the next method, and because the method may have subtasks, which are solved by submethods offering a functionality that is different from the assumptions of the subtask. Domain-independent methods use a method ontology, which, for example, might include concepts such as states, transitions, locations, moves, and constraints, whereas the user input and the (domain-specific) knowledge-base use a domain-oriented ontology, which might include concepts such as office workers, office rooms, and room-assignment constraints. Thus, a related issue to the configuration problem is the problem of mappings between ontologies (Gennari et al., 1994).In this paper, we shall address the configuration problem. The problems of how to organize a library of problem-solving methods and of how to select an appropriate method from such a library are beyond the scope of the paper. We shall introduce an approach for handling method ontologies when configuring a method from more elementary submethods. In our framework, configuration of a method means selecting appropriate submethods for solving the subtasks of which a method is composed. We introduce the notion of a subtask ontology in order to be able (i) to make the ontology of a method independent of the submethods that are chosen to solve its subtasks, and (ii) to specify how a selected submethod is adapted to its subtask environment. Our approach supports such a configuration process on multiple levels of subtasks and submethods. Furthermore, the access of domain knowledge is organized in such a way that no mapping of domain knowledge between the different subtask/submethod levels is required.The approach which is described in this paper provides a framework for handling some important aspects of the method configuration problem. However, our framework does not provide a complete solution to this problem. Furthermore, the proposed framework needs in the future a thorough practical evaluation by solving various application tasks.The rest of this paper is organized as follows. Section 2 provides a background to PROTÉGÉ-II, the board-game method, and the MIKE approach. Section 3 introduces the notions of method and subtask ontologies and discusses their relationships with the interface specification of methods and subtasks, respectively. Section 4 analyses the role of ontologies for the configuration of problem-solving methods and presents a model for configuring a problem-solving method from more elementary submethods that perform the subtasks of the given problem-solving method. In Sections 5 and 6, we discuss the results, and draw conclusions, respectively.2. Background: PROTÉGÉ-II, the Board-Game Method, and MIKEIn this section, we shall give a brief introduction into PROTÉGÉ-II and MIKE (Angele et al., 1996b) and will describe the board-game method (Eriksson et al., 1995) since this method will be used to illustrate our approach.2.1 Method Reuse for PROTÉGÉ-IIPROTÉGÉ-II (Puerta et al., 1992, Gennari et al., 1994, Eriksson et al., 1995) is a methodology and suite of tools for constructing knowledge-based systems. The PROTÉGÉ-II methodology emphasizes the reuse of components, including problem-solving methods, ontologies, and knowledge-bases.PROTÉGÉ-II allows developers to reuse library methods and to generate custom-tailored knowledge-acquisition tools from ontologies. Domain experts can then use these knowledge-Inputs ->Figure 2-1: Method-subtask decomposition in PROTÉGÉ-IIacquisition tools to create knowledge bases for the problem solvers. In addition to developing tool support for knowledge-based systems, PROTÉGÉ-II is also a research project aimed at understanding the reuse of problem-solving methods, and at alternative approaches to reuse.Naturally, the configuration of problem-solving methods for new tasks is a critical step in the reuse process, and an important research issue for environments such as PROTÉGÉ-II.The model of reuse for PROTÉGÉ-II includes the notion of a library of reusable problem-solving methods (PSMs) that perform tasks . PROTÉGÉ-II uses the term task to indicate the computations and inferences a method should perform in terms of its input and output. (Note that the term task is used sometimes in other contexts to indicate the overall role of an application system, or the application task .) In PROTÉGÉ-II problem-solving methods are decomposable into subtasks .Other methods, sometimes called submethods, can perform these subtasks. Primitive methods that cannot be decomposed further are called mechanisms . This decomposition of tasks into methods and mechanisms is shown graphically in Figure 2-1.Submethods and mechanisms should be reusable by developers as they build a solution to a particular problem. Thus, the developer should be able to select a generic method that performs a task, and then configure this method by selecting and substituting appropriate submethods and mechanisms to perform the method’s subtasks. Note that because the input-output requirements of tasks and subtasks often differ from the input-output assumptions of preexisting methods and mechanisms, we must introduce mappings among the task (or subtask) and method (or submethod)ontologies.PROTÉGÉ-II uses three major types of ontologies for defining various aspects of the knowledge-based system: domain , method , and application ontologies . Domain ontologies model concepts and relationships for a particular domain of interest. Ideally, these ontologies should be partititioned so as to separate those parts that may be more dependent on the problem-solving method. Method ontologies model concepts related to problem-solving methods, including input and output assumptions. To enable reuse, method ontologies should be domain-independent. In most situations, reusable domain and method ontologies by themselves are insufficient for a completeapplication system. Thus, PROTÉGÉ-II uses an application ontology that combines domain and method ontologies for a particular application. Application ontologies are used to generate domain-specific, method-specific knowledge-acquisition tools.The focus of this paper is on the configuration of problem-solving methods and submethods. Thus, we will describe method ontologies (see Section 3), rather than domain or application ontologies, and the mappings among tasks and methods that are necessary for method configuration (see Section 4).2.2 The Board-Game Method (BGM)We shall use the board-game method ((Eriksson et al., 1995), (Fensel et al., 1996a)) as a sample method to illustrate method configuration in the PROTÉGÉ-II framework. The basic idea behind the board-game method is that the method should provide a conceptual model of a board game where game pieces move between locations on a board (see Figure 2-2). The state of such a board game is defined by a set of assignments specifying which pieces are assigned to which locations. Developers can use this method to perform tasks that they can model as board-game problems.Figure 2-2: The board-game method provides a conceptual model where pieces movebetween locations.To configure the board-game method for a new game, the developer must define among other the pieces, locations, moves, the initial state and goal states of the game. The method operates by searching the space of legal moves, and by determining and applying the most promising moves until the game reaches a goal state. The major advantage of the board-game method is that the notion of a board game as the basis for the method configuration, makes the method convenient to reuse for the developer.We have used the board-game method in different configurations to perform several tasks. Examples of such tasks are the towers-of-Hanoi, the cannibals-and-missionaries, and the Sisyphus room-assignment (Linster, 1994) problem. By modeling other types of tasks as board games, the board-game method can perform tasks beyond simple games. The board-game method can perform the Sisyphus room-assignment task, for instance, if we (1) model the office workers as pieces, (2) start from a state where all the workers are at a location outside the building, and (3) move the workers one by one to appropriate rooms under the room-assignment constraints.2.3 The MIKE ApproachThe MIKE approach (Model-based and Incremental Knowledge Engineering) (Angele et al., 1996b) aims at providing a development method for knowledge-based systems covering all steps from knowledge acquisition to design and implementation. As part of the MIKE approach theKnowledge Acquisition and Representation Language KARL (Fensel et al., 1996c), (Fensel, 1995) has been developed. KARL is a formal and operational knowledge modeling language which can be used to formally specify a KADS like model of expertise (Schreiber et al., 1993). Such a model of expertise is split up into three layers:The domain layer contains the domain model with knowledge about concepts, their features, and their relationships. The inference layer contains a specifiation of the single inference steps as well as a specification of the knowledge roles which indicate in which way domain knowledge is used within the problem solving steps. In MIKE three types of knowledge roles are distinguished: Stores are used as containers which provide input data to inference actions or collect output data generated by inference actions. Views and terminators are used to connect the (generic) inference layer with the domain layer: Views provide means for delivering domain knowledge to inference actions and to transform the domain specific terminology into the generic PSM specific terminology. In an analogous way, terminators may be used to write the results of the problem solving process back to the domain layer and thus to reformulate the results in domain specific terms. The task layer contains a specification of the control flow for the inference steps as defined on the inference layer.For the remainder of the paper, it is important to know that in KARL a problem solving method is specified in a generic way on the inference and task layer of a model of expertise. A main characteristic of KARL is the integration of object orientation into a logical framework. KARL provides classes and predicates for specifying concepts and relationships, respectively. Furthermore, classes are characterized by single- and multi-valued attributes and are embedded in an is-a hierarchy. For all these modeling primitives, KARL offers corresponding graphical representations. Finally, sufficient and necessary constraints, which have to be met by class and predicate definitions, may be specified using first-order formulae.Currently, a new version of KARL is under development which among others will provide the notion of a method ontology and will provide primitives for specifying pre- and postconditions for a PSM (Angele et al., 1996a). Thus, this new version of KARL includes all the modeling primitives which are needed to formally describe the knowledge-level framework which shall be introduced in Sections 3 and 4. However, this formal specification is beyond the scope of this paper.3. Problem-Solving Method OntologiesWhen describing a PSM, various characteristic features may be identified, such as the input/output behavior or the knowledge assumptions on which the PSM is based (Fensel, 1995a), (Fensel et al., 1996b). In the context of this paper, we will consider a further characteristic aspect of a PSM: its ontological assumptions. These assumptions specify what kind of generic concepts and relationships are inherent for the given PSM. In the framework of PROTÉGÉ-II, these assumptions are captured in the method ontology (Gennari et al.,1994).Subsequently, we define the notions of a method ontology and of a subtask ontology, and discuss the relationship between the method ontology and the subtask ontologies associated with the subtasks of which the PSM is composed. For that discussion we assume that a PSM comes with an interface specification that describes which generic external knowledge roles (Fensel, 1995b) are used as input and output. Each role includes the definition of concepts and relationships for specifying the terminology used within the role.Fig. 3-1 shows the interface specification of the board-game method. We see, for instance, that knowledge about moves, preferences among moves, and applicability conditions for moves is provided by the input roles "Moves", "Preferred_Moves", and "Applic_Moves" (applicable moves), respectively; within the role "Moves" the concept "moves" is defined, whereas forexample within the role "Preferred_Moves" the relationship "prefer_m" is defined which specifies a preference relation between two moves for a given state (see Fig. 3-3).external knowledge role data flowFigure 3-1: The interface of the board-game methodSince the context in which the method will be used is not known in advance, one cannot specify which input knowledge is delivered from other tasks as output and which input knowledge has to be taken from the domain. Therefore, besides the output role "Solution", all other input roles are handled homogenously (that is, as external input knowledge roles).The interface description determines in which way a method can be adapted to its calling environment: a subset of the external input roles will be used later on as domain views (that is for defining the mapping between the domain-specific knowledge and the generic PSM knowledge).3.1 Method OntologiesWe first consider the situation that a complete PSM is given as a building block in the library of PSMs. In this case, a PSM comes with a top-level ontology, its method ontology, specifying all the generic concepts and relationships that are used by the PSM for providing its functionality. This method ontology is divided into two parts:(i) Global definitions, which include all generic concept and relationship definitions that are partof the interface specification of the PSM (that is, the external input and output knowledge roles of the PSM, respectively). For each concept or relationship definition, it is possible to indicate whether it is used as input or as output (however, that does not hold for subordinate concept definitions, i.e. concepts that are just used as range restrictions of concept attributes, or for high-level concepts that are just used for introducing attributes which are inherited bysubconcepts). Thus, the ontology specifies clearly which type of generic knowledge is expected as input, and which type of generic knowledge is provided as output.(ii) Internal definitions, which specify all concepts and relationships that are used for defining the dataflow within the PSM (that is, they are defined within stores).Within both parts, constraints can be specified for further restricting the defined terminology. It should be clear that the global definitions are exactly those definitions that specify the ontological assumptions that must be met for applying the PSM.We assume that a PSM that is stored in the library comes with an ontology description at two levels of refinement. First, a compact representation is given that just lists the names of the concepts and relationships of which the ontology is composed. This compact representation also includes the distinctions of global and internal definitions. It is used for providing an initial, not too detailed overview about the method ontology.Fig. 3-2 shows this compact representation of the board-game method ontology. We see that for instance "moves" is an input concept, "prefer_m" (preference of moves) is an input relationship, and "goal-states" is an output concept; "assignments" is an example of a subordinate concept which is used within state definitions, whereas "movable_objects" is an example of a high level concept. Properties of "movable_objects" are for instance inherited by the concept "pieces" (see below). As we will see later on, the concept "current-states" is part of the internal definitions, since it is used within the board game method for specifying the data flow between subtasks (see Section 3.2)Figure 3-2: The compact representation of the ontology of the board-game method Second, a complete specification of the method ontology is given. We use KARL for formally specifying such an ontology which provides all concept and relationship definitions as well as all constraints. Fig. 3-3 gives a graphic KARL representation of the board-game method ontology(not including constraints). We can see that for instance a move defines a new location for a given piece (single-valued attribute "new_assign" with domain "moves" and range "assignments") or that the preference between moves is state dependent (relationship "prefer_m"). The attribute "assign" is an example of a multi-valued attribute since states consist of a set of assignments.: is-aFigure: 3-3: The graphic KARL representation of the board-game method ontology When comparing the interface specification (Fig. 3-1) and the method ontology (Fig. 3-3) we can see that the union of the terminology of the external knowledge roles is equal to the set of global definitions being found in the method ontology.3.2 Subtask OntologiesIn general, within the PROTÉGÉ-II framework, a PSM is decomposed into several subtasks. Each subtask may be decomposed in turn again by choosing more elementary methods for solving it. We generalize this approach in the sense that, for trivial subtasks, we do not distinguish between the subtasks and the also trivial mechanisms for solving them. Instead, we use the notion of an elementary inference action (Fensel et al., 1996c). In the given context, such an elementary inference action may be interpreted as a "hardwired" mechanism for solving a subtask. Thus, for trivial subtasks, we can avoid the overhead that is needed for associating a subtask with its corresponding method (see below). That is, in general we assume that a method can be decomposed into subtasks and elementary inference actions.When specifying a PSM, a crucial design decision is the decomposition of the PSM into its top-level subtasks. Since subtasks provide the slots where more elementary methods can be plugged in, the type and number of subtasks determine in which way a PSM can be configured from otherdata flow subtaskinternal storeexternal knowledge rolemethods. As a consequence, the adaptability of a PSM is characterized by the knowledge roles of its interface description, and by its top-level task decomposition.For a complete description of the decomposition structure of a PSM, one also has to specify the interfaces of the subtasks and inference actions, as well as the data and control flow among these constituents. The interface of a subtask consists of knowledge roles, which are either external knowledge roles or (internal) stores , which handle the input/output from/to other subtasks and inference actions. Some of these aspects are described for the BGM in Figures 3-4 and 3-5,respectively.Figure 3-4: The top-level decomposition of the board-game methodinto elementary inference actions and subtasksFig. 3-4 shows the decomposition of the board-game method into top-level subtasks andFigure 3-5: The interface of the subtask "Apply_Moves"elementary inference actions. We can see two subtasks ("Apply_Moves" and "Select_Best_State") and three elementary inference actions ("Init_State", "Check_Goal_State", "Transfer_Solution"). This decomposition structure specifies clearly that the board-game method may be configured by selecting appropriate methods for solving the subtasks "Apply_Moves" and "Select_Best_State". In Fig. 3-5 the interface of the subtask "Apply_Moves" is shown. The interface specifies that "Apply_Moves" receives (board-game method) internal input from the store "Current_State" and delivers output to the store "Potential_Successor_States". Furthermore, three external knowledge roles provide the task- and/or domain-specific knowledge that is required for performing the subtask "Apply_Moves".Having introduced the notion of a method ontology, the problem arises how that method ontology can be made independent of the selection of (more elementary) methods for solving the subtasks of the PSM. The basic idea for getting rid of this problem is that each subtask is associated with a subtask ontology. The method ontology is then essentially derived from these subtask ontologies by combining the different subtask ontologies, and by introducing additional superconcepts, like for example the concept "movable_objects" (compare Fig. 3-2). Of course, the terminology associated with the elementary inference actions has to be considered in addition.Figure: 3-6: The method ontology and the related subtask ontologiesThe notion of subtask ontologies has two advantages:1) By building up the method ontology from the various subtask ontologies, the method ontologyis independent of the decision which submethod will be used for solving which subtask. Thus,a configuration independent ontology definition can be given for each PSM in the library.2) Subtask ontologies provide a context for mapping the ontologies of the submethods, which areselected to solve the subtasks, to the global method ontology (see Section 4).The type of mapping that is required between the subtask ontology and the ontology of the method used to solve the subtask is dependent on the distinction of input and output definitions within the subtask ontology (see Section 4). Therefore, we again separate the subtask ontology appropriately. In addition, a distinction is made between internal and external input/output. The feature "internal" is used for indicating that this part of the ontology is used within the method of which the subtask is a part (that is, for defining the terminology of the stores used for the data flow among the subtasks which corresponds to the internal-definitions part of the method ontology). The feature "external" indicates that input is received from the calling environment of the method of which the subtask is a part, or that output is delivered to that calling environment (that corresponds to the global-definitions part of the method ontology). This distinction can be made on the basis of the data dependencies defined among the various subtasks (see Fig. 3-6).In Fig. 3-7, we introduce the ontology for the subtask "Apply_Moves". According to the interface description which is given in Fig. 3-5, we assume that the current state is an internal input for "Apply_Moves", whereas the potential successor states are treated as internal output. Moves, preferences among moves, and applicability conditions for moves are handled as external input.Figure 3-7: The compact representation of the "Apply_Moves" subtask ontologyWhen investigating the interface and ontology specification of a subtask, one can easily recognize that even after having introduced the subtask decomposition, it is still open as to what kind of external knowledge is taken from the domain and what kind of knowledge is received as output from another task. That is, in a complex application task environment, in which the board-game method is used to solve, for example, a subtask st1, it depends on the calling environment of st1 whether, e.g., the preference of moves ("prefer_m") has to be defined in a mapping from the domain, or is just delivered as output from another subtask st2, which is called before st1 is called.4. Configuring Problem Solving Methods from More Elementary MethodsThe basic idea when building up a library of PSMs is that one does not simply store completely defined PSMs. In order to achieve more flexibility in adapting a PSM from the library to its task environment, concepts are required for configuring a PSM from more elementary building blocks (i.e., from more elementary methods (Puerta et al., 1992)). Besides being more flexible, such a configuration approach also provides means for reusing these building blocks in different contexts.Based on the structure introduced in Section 3, the configuration of a PSM from building blocks requires the selection of methods that are appropriate for solving the subtasks of the PSM. Since we do not consider the indexing and adaptation problem in this paper, we assume subsequently, that we have found a suitable (sub-)method for solving a given subtask, e.g. by exploiting appropriate semantic descriptions of the stored methods. Such semantic descriptions could, for instance, be pre-/postconditions which specify the functional behavior of a method (Fensel et al., 1996b). By using the new version of KARL (Angele et al., 1996a) such pre-/postconditions can be specified in a completely formal way.。

Prototype System for Knowledge Problem DefinitionAhmed M.Al-Ghassani,M.ASCE 1;John M.Kamara,M.ASCE 2;Chimay J.Anumba,M.ASCE 3;andPatricia M.Carrillo 4Abstract:Attitudes to knowledge management ͑KM ͒have changed considerably as organizations are now realizing its benefits.Imple-mentation,however,has been facing serious difficulties attributed either to not being able to anticipate the barriers when planning KM strategies or to using inappropriate methods and tools for implementation.These difficulties are more critical in construction due to the fragmented nature of the industry.This paper suggests that proper definition of a KM problem at the early stages of developing the KM initiatives will result in better control over the KM barriers.A methodology for identifying KM problems within a business context is then introduced.The methodology is encapsulated into a prototype software system,which facilitates its deployment in organizations and provides online help facilities.The methodology,development,operation,and evaluation of the prototype are described.The paper concludes that the prototype offers considerable potential for delivering a clarified KM problem and a distilled set of issues for an organization to address.This represents a significant first step in any KM initiative.DOI:10.1061/͑ASCE ͒0733-9364͑2006͒132:5͑516͒CE Database subject headings:Information technology ͑IT ͒;Information management;Knowledge-based systems;Construction industry .IntroductionKnowledge relates to data and information and this relationship needs to be understood first.Data can be described as discrete facts about events which when processed and given relevant as-sociations and patterns become information ͑Blumentritt and Johnston 1999͒.Knowledge is therefore information with context that provides the basis for actions and decision making ͑Kanter 1999͒.There are several definitions of knowledge management ͑KM ͒.It can be defined from a process perspective,an outcome perspec-tive,or a combination.An example of a process perspective definition sees KM as a process of controlling the creation,dis-semination,and utilization of knowledge ͑Newman 1991;Kazi et al.1999͒.Another considers KM as the “identification,optimiza-tion,and active management of intellectual assets,either in the form of explicit knowledge held in artifacts or as tacit knowledgepossessed by individuals or communities to hold,share,and grow the tacit knowledge”͑Snowden 1998͒.The outcome perspective,on the other hand,focuses on the benefits that an organization gets from managing its knowledge.An example is a definition by Kanter ͑1999͒,who sees KM to be concerned with the way an organization gains competitive advantage and builds an innova-tive and successful organization.Another example considers KM as the “management of organizational knowledge for creating business value and generating competitive advantage”͑Tiwana 2000͒.A third example defines KM as “the ability to create and retain greater value from core business competencies”͑Klasson 1999͒.A combined definition describes both the process by which knowledge is managed and the outcome that is likely to result.Tiwana ͑2000͒,for example,states that “Knowledge management enables the creation,communication,and application of knowl-edge of all kinds to achieve business goals.”Another definition by Scarbrough et al.͑1999͒states that KM is any process or practice of creating,acquiring,capturing,sharing,and using knowledge,wherever it resides,to enhance learning and performance in or-ganizations.Regardless of the different perspectives for defining KM,all definitions focus on the fact that knowledge is a valuable asset that must be managed,and that KM is important to provide strategies to retain knowledge and to improve performance.The concept of KM is better understood when compared to other concepts such as the organization of work,learning organi-zation,the management of people,sources of wealth creation in contemporary society,the emergence of the information age,or-ganizational learning,the strategic management of core compe-tencies,the management of knowledge-intensive firms,the value of intellectual capital,and the management of research and devel-opment ͑Scarbrough et al.1999͒.For example,the information age is more broadly focused and affects all firms while the man-agement of research and development is more narrowly focused and is relevant to a limited number of firms.On the other hand,KM is more narrowly focused on the ways in which firms facing highly turbulent environments can mobilize their knowledge base1Assistant Dean for Academic Affairs,Al-Musanna College of Technology,P.O.Box 191,P.C.314,Al-Muluddah,Oman.E-mail:ahmed@ 2Senior Lecturer,School of Architecture,Planning and Landscape,Univ.of Newcastle upon Tyne,3-4Claremont Terrace,Newcastle upon Tyne NE24AE,U.K.E-mail:j.m.kamara@ 3Professor,Dept.of Civil and Building Engineering,Loughborough Univ.,Loughborough,Leicestershire,LE113TU,U.K.E-mail:c.j.anumba@ 4Professor,Dept.of Civil and Building Engineering,Loughborough Univ.,Loughborough,Leicestershire,LE113TU,U.K.E-mail:p.m.carrillo@Note.Discussion open until October 1,2006.Separate discussions must be submitted for individual papers.To extend the closing date by one month,a written request must be filed with the ASCE Managing Editor.The manuscript for this paper was submitted for review and pos-sible publication on January 24,2002;approved on May 18,2005.This paper is part of the Journal of Construction Engineering and Manage-ment ,V ol.132,No.5,May 1,2006.©ASCE,ISSN 0733-9364/2006/5-516–524/$25.00.D o w n l o a d e d f r o m a s c e l i b r a r y .o r g b y W u h a n U n i v e r s i t y o n 11/13/13. C o p y r i g h t A S CE .F o r p e r s o n a l u s e o n l y ; a l l r i g h t s r e s e r v e d .͑or knowledge assets ͒in order to ensure continuous innovation in projects ͑Scarbrough et al.1999͒.This paper presents a methodology developed for identifying KM problems,within a business context,for construction and manufacturing organizations.It then describes the development,operation,and evaluation of a prototype system produced to en-capsulate this methodology.The paper first reviews current KM practices.Current Knowledge Management Practices Overview of Research SurveysKnowledge management is making its mark on organizations of all sizes and in all sectors.Surveys show that the attitude to KM has changed since 1998͑KPMG 1998͒.This change in attitude had several causes,mainly the successful implementation of KM in some organizations and the publishing of two important books on KM by Nonaka and Takeuchi ͑1995͒and Davenport and Pru-sak ͑1998͒.A survey by the Information Systems Research Center of Cranfield School of Management in September 1997found that almost one-third of respondents thought that KM was just a pass-ing fad ͑KPMG 1998͒.Surveys after 1998show that there is a strong belief in the benefits of KM and that it is no longer seen as a passing fad ͑TFPL 1999;Gottschalk 1999;Robinson et al.2001͒.The results of a number of surveys that investigated the awareness of KM and the breadth with which it is implemented are discussed below.A survey of leading anizations representing different industry sectors with turnover exceeding £200million ͑$360mil-lion ͒a year was undertaken by KPMG Management Consulting ͑KPMG 1998͒.The results of the 100respondents show that KM is not seen as a fad any more but increasingly taken seriously.This was confirmed by 43%of the respondents who considered their organizations to have KM initiatives in place.The survey also shows that the awareness of KM increases with the size of the organization.It also informs us that some organizations imple-menting KM have already seen real benefits while others had difficulties.Another survey by TFPL Ltd.covered 500organizations implementing KM or equivalent initiatives from all business sec-tors around the world ͑TFPL 1999͒.Results show that users be-lieve that the concept of KM relies on other concepts such as succeeding in business and enabling organizations to meet their corporate objectives.Of the 80respondents,29%had corporate-wide KM programs and 18%were planning a corporate-wide KM program.50%had no corporate-wide KM program,but of them,42%had another corporate program with similar objectives.The survey concludes that the level of interest in KM and the number of organizations implementing its initiatives was growing expo-nentially.Moreover,many chief executives placed KM as second on their list of “must-dos”after globalization.The survey also shows that most of the KM literature before 1998concentrated on selling the concept of KM using examples of a few innovative adopters and that much of the focus was on demonstrating the value of intellectual capital,the leveraging of knowledge for com-petitive advantage,structural issues that underpin KM,and that KM is mostly about people and culture.Other surveys that focused on particular industry sectors show similar results.A survey covering 73respondents from 256Nor-wegian law firms and interviews with ten of the largest firms shows that there was a strong belief in the potential benefits ofKM ͑Gottschalk 1999͒.Another survey covering 170construction organizations ͑consultants and contractors ͒in the United Kingdom shows that about 40%already had a KM strategy,an-other 41%had plans to have a strategy within a year,and 19%did not have a strategy ͑Robinson et al.2001͒.The research also found that about 50%of U.K.construction organizations had al-ready appointed a knowledge manager or a special group with responsibility for implementing their KM strategy.It is therefore agreed that KM is an important concept that attracts an increasing number of organizations and that those or-ganizations that implement it first will gain more benefits.Tiwana ͑2000͒believes that organizations which decide to wait until KM becomes a hot issue are likely to fall behind,perhaps never to recover.Knowledge Management in ConstructionAlthough the term “KM”is relatively new to construction organi-sations ͑Carrillo et al.2000͒,many have adopted strategies for its implementation because the design and construction of projects involve several knowledge-intensive activities.These activities ͑e.g.,preliminary design,analysis,detailed design,planning and managing the construction,maintenance,etc.͒are influenced by factors that are linked to human intelligence and knowledge,such as experience and engineering judgment.The availability of knowledge is an important factor that affects construction projects as even experienced staff face difficulties when the required knowledge is not available,and this can result in assumptions or judgments that may be disproved when knowledge becomes available ͑Kumar and Topping 1991͒.Construction knowledge takes many forms,e.g.,experiences,best practices,lessons learned,drawings,documents,etc.This knowledge is developed within design offices or on construction sites.Knowledge developed within a design office is easy to ac-cess by individuals or groups working within the same office but those located in geographically dispersed offices will not have the same ease of access to this knowledge.Knowledge generated on a construction site is rarely shared and this can result in loss of this knowledge.For example,if a problem relevant to the perfor-mance of a structure occurs on a construction site,engineers in the design office need to know about this problem ͑its nature,why it occurred,and how it was solved ͒so that they do not repeat mistakes in future designs.Improving the “knowledge flow”within an organization adds value,increases the ability to com-pete,and helps to improve the quality of future projects.Construction organizations can benefit from KM by imple-menting initiatives that help in capturing knowledge that is gen-erated during the different stages of a project life cycle to make it available and accessible in a timely fashion,throughout the orga-nization.Managing construction knowledge not only contributes to increased safety and improved stability but also saves time spent in design and construction and provides scope for innova-tion.Time can be saved,for example,by reducing the number of design cycles or by reducing the time used for searching for knowledge.Therefore,there is need for an approach that facili-tates knowledge sharing within the construction industry.Barriers to Knowledge Management Implementation All the aforementioned surveys agree that organizations imple-menting KM face some barriers.The presence and strength of these barriers are dependent on many factors,such as the type of business processes,products,and clients.Construction organiza-D o w n l o a d e d f r o m a s c e l i b r a r y .o r g b y W u h a n U n i v e r s i t y o n 11/13/13. C o p y r i g h t A S CE .F o r p e r s o n a l u s e o n l y ; a l l r i g h t s r e s e r v e d .tions need to carefully investigate the barriers that are more rel-evant to construction activities in order to develop reliable KM strategies.Long lists of barriers are identified in the literature.However,they can be categorized into three main “barrier groups:”status of knowledge,location of knowledge,and culture surrounding the knowledge.Other barriers,such as the high cost of KM systems,are not considered as main barriers because they are easier to address,e.g.,by allocating a budget for implement-ing KM.The main barrier groups are discussed in turn below.Status of KnowledgeKnowledge exists in a tacit or explicit status ͑Nonaka and Takeuchi 1995͒and the allowance for an easy conversion from one status to another is important.Each status has its own char-acteristics,which may support or resist the conversion process.Knowledge,whether tacit or explicit,is either fully developed or still developing and this can also affect the conversion process as will be discussed below.Tacit Knowledge.This is stored in peoples’brains as mental models,experiences,and skills and is difficult to communicate externally ͑Vail 1999͒.The conversion of tacit knowledge takes two forms ͑Nonaka and Takeuchi 1995͒.It can be converted to other tacit knowledge through socialization in face-to-face inter-actions or to explicit knowledge through externalization by codi-fying an individual’s knowledge.Capturing tacit knowledge and codifying it is one of the biggest challenges in KM ͑Bair and O’Connor 1998͒.This knowledge,whether still “developing”͑in ongoing projects ͒or already “developed”͑in previous projects ͒,requires methodologies to help manage it wherever it exists.Cap-turing the developing knowledge during the life cycle of a project creates additional tasks in the congested agenda of employees.One of the evolving solutions is to integrate KM into the organi-zational processes.However,this could lead to several delays in finishing the work on time due to added activities.Furthermore,capturing the developed tacit knowledge is far more difficult and requires advanced tools and strategies.Explicit Knowledge.This is encoded in organizational formal models,rules,documents,drawings,products,services,facilities,systems,and processes and is easily communicated externally ͑Vail 1999͒.Its conversion also takes two forms ͑Nonaka and Takeuchi 1995͒.It can be converted to tacit knowledge through internalization when an individual reads and understands well-coded knowledge.It can also be converted to another type of explicit knowledge through combining more than one form of knowledge to generate new knowledge.Although conversion of explicit knowledge is easier than that of tacit knowledge,it still requires several resources such as time,technology,and commit-ment.Many organizations find that developed explicit knowledge is more difficult to manage than developing knowledge because it is hard to manage knowledge that was developed ͑in the past ͒by other individuals.In the long term,however,developed knowl-edge is more critical to the organization’s knowledge base as it constitutes the memory of the organization.The difficulty associ-ated with managing developed knowledge is with regard to allo-cating resources,especially time and people to find and store this knowledge.Location of Knowledge and Its Intended UsersKnowledge,whether tacit or explicit,exists in different sources and is required by different users.A source of knowledge can be a user of another knowledge,for example a person is a sourcewhen he contributes to a knowledge base but a user when he reads from the knowledge base.The transformation of knowledge from sources to users is facilitated or resisted by several factors.Understanding the relationships between these sources and users and their associated enablers and resistors is a difficult task re-quiring extensive examination.Moreover,it becomes more diffi-cult to manage knowledge when its sources or users are geo-graphically dispersed.Geographically Dispersed Sources.In the construction con-text,these include:offices,employees,designers,contractors,suppliers,customers,and partners.The coordination of the pro-cess of capturing knowledge from geographically dispersed sources is obstructed by many factors.For example,capturing knowledge from people spending most of their time on construc-tion sites is opposed by limited time and difficulty in getting technologies for capturing knowledge into the sites.Furthermore,having dispersed sources of knowledge ͑e.g.,staff scattered in different countries ͒requires a central unit that plans and monitors the process to ensure that correct and valid knowledge is collected ͑Conheeney et al.2000͒.Geographically Dispersed anizations with geo-graphically dispersed offices and employees need to ensure that the captured knowledge is safely delivered to users whenever they require it.Making this knowledge available to its intended users has several challenges with regard to the methodology of trans-ferring and sharing,supporting technology,and availability of time for accessing the transferred knowledge ͑Patel et al.2000͒.CultureCultural barriers that obstruct the implementation of KM refer to the culture of individuals and organizational culture.Culture of individuals is about the way individuals act and respond.For example,not all employees in an organization will have the same willingness to share their anizational culture is more about the processes used within the organization.For ex-ample,some organizations consider review meeting as part of the design process and this encourages knowledge sharing.The cul-tural barriers that affect the implementation of KM are given below.Willingness to Share.“Why would I give my knowledge to others?”This is a typical question that is asked by people when they are asked to share their knowledge.Most of the literature focuses on the organizational benefits of KM,neglecting the fact that tacit knowledge cannot be captured unless its holders realize that they also benefit ͑Scarbrough et al.1999͒.Employees need to understand that shared knowledge multiplies and that everyone needs knowledge from others.Second,people like to be rewarded for their contribution to the organizational knowledge base and this necessitates the adoption of carefully developed reward sys-tems that do not focus only on monetary figures but also include other schemes such as promotion,recognition,etc.Availability of Time.Employees find themselves under pres-sure of increased job tasks and delivery deadlines.Codifying tacit knowledge is difficult and time consuming and even storing and indexing codified knowledge is time intensive.Many organiza-tions therefore find difficulties in allocating time to their staff to contribute their knowledge and to manage it.Employees also need time for training in order to understand the KM system so that they can use it efficiently.Employees who need to search forD o w n l o a d e d f r o m a s c e l i b r a r y .o r g b y W u h a n U n i v e r s i t y o n 11/13/13. C o p y r i g h t A S CE .F o r p e r s o n a l u s e o n l y ; a l l r i g h t s r e s e r v e d .an answer to a question also find that they do not have enough time to search the knowledge base.Instead,they normally prefer to ask an experienced colleague ͑Carrillo et al.2000͒.Type and Nature of anizations differ according to the industry sector within which they perform.Some organiza-tions have most of their work indoors,e.g.,design offices,while others have most of their work outdoors,e.g.,civil engineering projects.Capturing knowledge from outdoor projects is very dif-ficult due to two types of mobility.One is with regard to the formation of new teams for each project and the second is with regard to mobility of staff within the project site.Some project sites do not have a suitable environment for capturing knowledge or even for accessing a knowledge base.Technology Infrastructure.Although technology is only a facilitator to KM,many of the KM processes depend on it to allow for faster storage,retrieval,and transfer ͑Tiwana 2000͒.Technology therefore provides support for KM with a range of hardware and software tools.However,many organizations find it difficult to identify the tools that address their needs since this requires an understanding of the KM requirements for the orga-nization and what these tools can ing the wrong tools can result in a technology infrastructure that is not compatible with the existing technology within the organization or that does not address the organization’s business goals.Size of Organization.The amount of knowledge available in an organization may be considered proportional to the size of the organization.Knowledge within a large organization is scattered throughout its offices and is therefore more difficult to manage.This necessitates that organizations identify what knowledge they need to manage and where it exists in order to achieve the orga-nizational business goals ͑CIRIA 2000͒.This identification may not be a straightforward process as managing more than one knowledge type may be required.In this case,an organization needs to prioritize these types of knowledge based on the priori-ties of its business goals.Rewarding Schemes May Also Create a Barrier.Many KM initiatives focus on some kind of rewarding schemes,especially short-term schemes.The problem with these schemes is that they consider rewarding people for what they do.This can create a new barrier because people may become more materialistic andwill not accept new work unless they are explicitly rewarded for doing it.Therefore reward schemes need to be carefully designed so that they do not create a new barrier.Overcoming the BarriersLiterature and surveys show that many barriers obstruct the implementation of KM ͑Davenport 1997;KPMG 1998;Gottschalk 1999;TFPL 1999;Scarbrough et al.1999;Carrillo et al.2000;CIRIA 2000;Patel et al.2000;Storey and Barnet 2000;Tiwana 2000;Robinson et al.2001͒.These barriers,if not prop-erly addressed,can result in losing trust in the concept of KM ͑McConalogue 1999;Storey and Barnet 2000͒.New barriers can also result from poor practices or badly designed systems.Over-coming KM barriers is not an easy task and requires extensive planning ͑Al-Ghassani et al.2001b ͒.This necessitates fully un-derstanding the nature of knowledge that needs to be managed and the barriers that resist its implementation.One of the best ways for understanding existing barriers is by having clear iden-tification of KM problems at the early stages of developing strat-egies for KM ͑Al-Ghassani et al.2000b ͒.Early identification of KM problems is crucial because systems are more flexible during their design phase,while rectifying or altering a KM system at a later stage is difficult if at all possible,time consuming,and ex-pensive ͑CIRIA 2000͒.Organizations therefore need a tool that helps capture and define the KM problem so that reliable KM initiatives can be developed.Although numerous KM tools exist to support some elements of the KM process ͑Ruggles 1997;Jackson 1998;Bair and O’Connor 1999;Tiwana 2000;Al-Ghassani et al.2001a;Carrillo et al.2000͒,no tool is found to facilitate understanding and defining the KM problem.Methodology for Identifying Knowledge Management Problems BackgroundDefining a KM problem requires intensive examination of several issues.A methodology for identifying KM problems is therefore developed within the cross-sectoral learning in the virtual enter-prise research ͑CLEVER ͒project ͑Kamara et al.2001͒.The CLEVER framework ͑Fig.1͒is developed to support KM in con-struction and manufacturing organizations.It aims to clarifyaFig.1.The CLEVER framework for knowledge managementD o w n l o a d e d f r o m a s c e l i b r a r y .o r g b y W u h a n U n i v e r s i t y o n 11/13/13. C o p y r i g h t A S CE .F o r p e r s o n a l u s e o n l y ; a l l r i g h t s r e s e r v e d .vague KM problem into a set of specific KM issues,established within a business context in order to provide appropriate and relevant processes to solve the identified problems by•Defining the KM problem and linking it to business drivers or goals;•Creating the desired characteristics of KM solutions;•Identifying the critical migration paths to achieve the desired characteristics;•Selecting appropriate KM processes to use on those paths;and •Developing a strategy for implementing the selected KM processes.The framework addresses these objectives through the four main stages illustrated in Fig.1.The first stage,“Identify KM Problem,”aims to clarify the overall KM problem within a busi-ness context to deliver a refined KM problem and a set of KM issues from the overall problem.The second stage,“Identify Cur-rent and Required KM Characteristics,”introduces a series of knowledge dimensions, e.g.,explicit/tacit,critical/auxiliary,generic/project specific,etc.This stage asks users to identify cer-tain characteristics about the knowledge that they are interested in managing.Following this,users are asked to identify existing knowledge characteristics and the desired knowledge characteris-tics.The third stage,“Identify Critical Knowledge Migration Paths,”aims to allow users to select how best they would like to migrate from the existing characteristics to the desired character-istics,i.e.,either directly,or using intermediate steps.The last stage,“Select Generic KM Processes,”aims to ensure that the organization is in a position to implement KM and helps in se-lecting the appropriate KM processes which,when tailored to a particular organization’s need,will address the stated KM problem.The subsequent sections discuss,in some detail,the method-ology,development,and operation of a prototype developed to address the first stage “Identify KM Problem”through introduc-ing the problem definition template ͑PDT ͒.The Problem Definition TemplateThe problem definition template represents a methodology devel-oped for clarifying the overall KM problem within an organiza-tional business context.It aims to assist users to “think through”the problem in a “structured”way.It therefore covers issues that can cause,if not properly addressed,a weak definition of KM problems.These issues are identified by Al-Ghassani et al.͑2004͒as improper identification of the type and nature of knowledge that needs to be managed;unclear business goals from imple-menting KM initiatives;improper identification of the character-istics of knowledge;and poor understanding of the relationships between sources and users of knowledge and their associated en-ablers and resistors.In order to address these issues,the process consists of several stages,each comprising a set of investigations that address relevant KM issues.The developed approach allows organizations to•describe a “vague”KM problem which does not need to be too specific at this stage;•investigate the problem against the organizational business drivers;•characterize the knowledge problem;•identify the sources and users of the knowledge;•identify the enablers and resistors for transferring the knowl-edge from sources to users;•link the knowledge problem to the relevant KM process ͑es ͒;and finally•refine the previously stated KM problem.Using this approach allows organizations to identify if their business drivers have a KM dimension and enables them to de-velop strategies to address the KM problem.This new approach of gathering information for the identification of KM problems is simple to use and cost-effective.It is applicable to large organi-zations as well as small-to-medium-sized organizations,whichFig.2.Section of the paper version of problem definition templateD o w n l o a d e d f r o m a s c e l i b r a r y .o r g b y W u h a n U n i v e r s i t y o n 11/13/13. C o p y r i g h t A S CE .F o r p e r s o n a l u s e o n l y ; a l l r i g h t s r e s e r v e d .。

Building a Large Knowledge Base from a Structured Source:The CIA World Fact BookGleb FrankAdam FarquharRichard Fikes{frank,farquhar,fikes}@Knowledge Systems LaboratoryComputer Science DepartmentStanford UniversityAbstractThe on-line world is populated by an increasing number of knowledge-rich resources. Furthermore, there is a growing trend among authors to provide semantic markup of these resources. This presents a tantaliz-ing prospect. Perhaps we can leverage the person-years of effort invested in building these knowledge-rich resources to create large-scale knowledge bases.The World Fact Book knowledge base has been an experiment in the construction of a large-scale knowl-edge base from a source authored using semantic markup. The content of the knowledge base is, in large part, derived from the CIA World Fact Book, and covers a broad range of information about the world’s nations. The World Fact Book is a highly structured document with a complex underlying ontology. The structure makes it possible to parse the document in order to carry out the knowledge extraction. However, irregularities of the text written by humans and the complexity of the domain make the knowledge extrac-tion process non-trivial.We describe the process we used to construct the World Fact Book knowledge base, including parsing the source, refining the implicit knowledge, constructing a substantial supporting ontology, and reusing exist-ing ontologies. We also discuss some of the key representational issues addressed and show how the re-sulting axioms can be used to answer a variety of queries.We hope that the broad accessibility of the resulting knowledge base and its neutral representational format will enable others to work with and extend the content, as well as explore issues of structuring and infer-encing in large-scale knowledge bases.1 IntroductionThe on-line world is being populated by an increasing number of knowledge rich sources including dictionaries, encyclopedias, and documents with specialized information on many domains. Furthermore, there is a growing trend to provide these sources in a highly structured form that includes both syntactic markup, in a language such as the hypertext markup language (HTML), and semantic markup, in a language such as the extensible markup lan-guage (XML or SGML). These documents are a critical resource for the construction of the next generation of large scale knowledge bases. They provide the opportunity to leverage many person-years of effort to create useful high-quality knowledge bases in a variety of domains. Of course, these documents are not, in themselves, knowledge bases. They use natural language fragments, have many irregularities in their structure, and rely on the reader to provide substantial amounts of background knowledge to interpret their content. We must overcome these barriers in order to exploit this potentially valuable knowledge work.This paper describes an effort to build a very large-scale knowledge base (KB) from a knowledge rich source. The content of the knowledge base derives from the CIA’s World Fact Book (WFB), which is a collection of geographi-cal, economic, sociological and other facts about the countries and territories of the world [1]. The version of the WFB that we used is specified using SGML, which makes substantial portions of its content accessible. We have converted and refined the textual content of the WFB into an extensive knowledge base with an underlying formal ontology. The result is a computer-usable very large-scale knowledge base that, besides being an artifact of intrinsic interest, can be used as a testbed for exploring the scalability of reasoning tools and problem solving architectures.1Much of the research in Knowledge Representation and Reasoning has been organized around the solution of small canonical problems such as the Yale shooting problem, the Nixon diamond, cascading bathtubs, and so on. A good canonical problem encapsulates the critical issues that must be addressed in order to solve an important class of real world problems. There are, however, important classes of problems whose answers can only be empirically deter-mined by working with large knowledge bases. For example, we would like to understand how to structure many thousands of classes in order to support efficient inference and human understanding; we want to develop efficient inference methods in the presence of many thousands of irrelevant axioms. Many current knowledge representation systems are effectively incapable of working with knowledge bases of the scale of the WFB KB. We hope that the WFB KB can serve as a substrate upon which these and other empirical questions can be resolved. We think that the use of large-scale knowledge bases such as the WFB KB can help to close the gap between small canonical problems and the real world. Two notable efforts that have developed large scale knowledge bases are the botany knowledge base project [2] and the CYC project [3].The WFB KB is also an experiment in knowledge reuse. We have reused definitions and axioms from the Ontolin-gua library [4], as well as the Upper Level ontology that has been developed within the High Performance Knowl-edge Base (HPKB) project of the Defense Advanced Research Projects Agency. This reduces the task of designing equivalent representations and increases the likelihood of interoperation with other knowledge bases. Knowledge reuse experiments also provide a foundation for current research on establishing the better ways to structure and combine ontologies.1.1 ArchitectureThe process of creating the WFB KB consisted of three stages. The first stage was knowledge extraction. During the knowledge extraction stage, knowledge was extracted from the SGML source using a custom parser written spe-cifically for this purpose. The second stage was knowledge refinement. During the knowledge refinement stage, many of the terms identified in the first phase were organized and linked into taxonomies. The third stage was on-tology construction. During the ontology construction phase, new classes and axioms were written that expand the definitions introduced during the knowledge refinement stage. The facts extracted during the first stage were also connected by lifting axioms to the ontology.The first stage was largely automatic and consisted of using a parser to transliterate the SGML document into rela-tional sentences in the knowledge interchange format (KIF) [9]. The parser generated two major types of informa-tion: sentences expressing the explicit facts from the WFB and terminological information about the constants used in the sentences expressing the facts. For example, a WFB entry might express the explicit fact that "Oil makes up 90% of the exports of Saudi Arabia" and thereby imply the terminological information that "Oil is a trading com-modity". The terms described in the terminological information are those that appear in the WFB. No attempt was made in this stage to translate those terms into the terminology of existing ontologies.The parser we used to perform this knowledge extraction also produced files that contained all of the data values from the SGML input file; an exceptions file that contained those irregular entries that could not be parsed; and a “source” file that contained the parsed entries in the following format:(wfb::source-text <field> <country> [<subfield>] “<data>”),e.g.,(wfb::source-text exports haiti partners "US 81%, Europe 12% (1993)" ) The second stage of the knowledge extraction process consisted of producing an ontology for representing the im-plicit facts and was done by hand. The flat lists of terms of a particular type that the parser had generated (e.g., lists of natural resources and languages) were transformed into taxonomies. We also represented obvious functional connections between terms of different types, e.g., between a natural resource and an industry that exploits it or a language and an ethnic entity that speaks it. We then organized and modularized these taxonomies, and connected the classes and relations in them to the corresponding super-classes and super-relations in the HPKB upper level ontology. Finally, we did the difficult and large task of designing ontologies that could be used to express the WFB facts in standard vocabularies and in a form that enables their effective use by general-purpose reasoners.The third stage of the extraction process consisted of writing new axioms covering the terms introduced during re-finement and lifting axioms that transform the KIF sentences produced by the parser in the first stage into sentences expressed in these ontologies. Instead of transforming the parser output into a new format, we decided to keep the KIF files produced by the knowledge extraction process, and connect them to the target OKBC-compliant [6] ontol-ogy through lifting axioms. The facts are kept in the format of the parser output. The queries are formulated in the2terms of the target frame-language ontology. They are then reformulated into the terms of the facts file through the corresponding lifting axiom.Figure 1: The diagram shows the flow of information during the extraction process. Solid black arrows indicate parsing. The block arrow from terms to taxonomies indicates the manual construction of taxonomies from flat lists of terms. Dashed arrows indicate runtime information flow.2 Knowledge ExtractionKnowledge extraction from the WFB is possible because it is a highly structured document. The WFB is available in several different forms: as an HTML document, as a text document, as an interactive web service, and in SGML. Of these forms, the SGML provides both the most regular syntactic structure as well as the most semantic markup. Although the SGML version that we used is not the most current (we had access to the 1995 WFB in SGML), its advantages outweighed any disadvantages, so we chose it as the basis for our work. Our primary interest was the task of representing the knowledge in the WFB, rather than assuring the currency and accuracy of the information contained therein.Knowledge extraction from the WFB is possible because it is a highly structured document. The information about countries is organized alphabetically. The facts about each country are structured into fields and, sometimes, sub-fields. The fields and subfields are almost completely regular, that is all countries have almost identical sets of fields and subfields.To each field (and subfield, if it is present) corresponds a data value, which is a numeric value, a fragment of Eng-lish text, or a list of data values. Unfortunately, the format is different for almost all fields, so that data values for each field must be parsed in a separate way. Furthermore, the format of data values is often quite irregular even within a given field, and the same type of information about different countries may be presented in different ways.A typical entry in the WFB within the section about Italy can look like this:<field>Population growth rate</field><data>0.21%</data>.This is parsed into the KIF formula:(wfb::population-growth-rate italy (* 0.21 %) )We tried to extract as much information as we could without involving serious natural language processing tech-niques. The amount and quality of information therefore depended on the regularity of the format of the correspon-dent data values in the input document. Those fields whose data values were unstructured English text, such as the3descriptions of climate and terrain, and those fields whose data values were extremely irregular were not parsed. The data values in such cases were simply left as English text fragments.Even those fields that were structured and regular enough to be parsed often were not uniform throughout the input file. If the majority of the data values for a particular field displayed a particular structure, and only several entries were constructed in a different way, these abnormal entries were left unparsed and were redirected in a special "ex-ceptions" file. Many of these records just say that the information is not available.When the information in the data values described an object with a relation to a country (e.g., a city), or a class of such objects (e.g., coal deposits), the parser created a symbolic constant for the object or class. That is, the symbolic constant, as opposed to the string, was used in the corresponding KIF fact and the appropriate taxonomic informa-tion was placed in the terms file. Extra attention was necessary to prevent name clashes. For example, the word “textile” appears both as a commodity and as an industry. The method of solving name clashes depended on the context in which the objects appeared. For example, if two countries both had a natural resource called "coal", it was represented by a single constant. When two countries had a city called Santiago, however, two distinct constants were created.Another problem of object creation was the existence of synonyms. Some objects in the WFB are referred to by several distinct names (e.g., US, USA, United States, and United States of America), and others are close semantic neighbors (e.g., wolfram and tungsten). Most of the synonymy relationships cannot be established at the parsing stage. As a consequence, the problem of synonyms and related problems are resolved during the knowledge re-finement stage.In the case that a data value specifies a list of items (e.g, for a nation's industries or natural resources), it may be dif-ficult to decompose the list into its constituents. For example, the value “small deposits of coal and natural gas”means “small deposits of coal” and “small deposits of natural gas”, whereas “coal and natural gas” means “coal” and “natural gas”. Such problematic cases introduce a single constant, which is decomposed manually during the knowledge refinement stage.3 Knowledge RefinementThe raw facts contained in the WFB can be quite useful by themselves, however they also provide an opportunity for reasoning that is much more powerful. The knowledge extraction process produces terms, organized into other-wise unstructured groups. For example, the group of industries includes the terms manufacturing and automobile manufacturing. If a group contains individuals, then the group will correspond to a class in the target ontology (e.g., cities), otherwise it will correspond to a metaclass (e.g., industries). Clearly, if we brought outside knowledge to bear on these term groups, we could define a richer structure that would support stronger inference.While there are thousands of terms in the WFB, the number of terms in each group is much more limited. A typical group will include less than 500 terms. This makes it possible to organize the terms by hand. This is the primary task of knowledge refinement. The knowledge refinement stage, however, involves three subtasks. First, synonyms are eliminated through the use of simple rewrite rules that establish a preferred term in each set of synonyms (e.g.,(preferred-term oil petroleum)). Second, the remaining unique terms are organized into a taxon-omy. This allows a system to answer more concrete queries (e.g., “What fossil fuels are there in Saudi Arabia?” as opposed to “What natural resources are there in Saudi Arabia?”). It also allows a system to answer more general queries (e.g., “Are there any fossil fuels in Saudi Arabia?” as opposed to “Is there any petroleum in Saudi Arabia?”). Finally, the list elements that the parser was unable to decompose are manually split.3.1 Taxonomy constructionThe parser emits several types of objects for which meaningful taxonomies can be built. They are industries, com-modities, natural resources, languages, religions and ethnic groups. The information how to taxonomize these ob-jects is not in the WFB; at this stage we relied exclusively on outside knowledge.Many of the classes in the WFB are non-primitive. That is, they are defined by sufficient conditions. The obvious reason for that is that these classes were originally introduced by people, they are not “natural” classes. When the authors of the WFB associated these classes with countries, they based their decisions on some implicit criteria. For example, the ground for the statement that India’s industrial sectors include machinery production is probably that there are a significant number of businesses in India that can be described as producing machinery. These rules are sufficient conditions of membership in the class of machinery-production businesses. Therefore, the machinery in-dustrial sector, viewed as a class of businesses, is a non-primitive class. Many of the classes in the WFB can be4naturally organized into taxonomies according to multiple attributes. Oil resources, for example, can be classified into offshore and inland deposits, according to the size and significance of deposit, or how fully they have been ex-ploited. One solution is to classify objects along several orthogonal dimensions, introduce classes for each such dimension, and then subclass each WFB class from the appropriate dimensions. For example, the WFB class "small unexploited deposits of iron" is a subclass of iron deposits, of low-abundance natural resources, and of unexploited natural resources.Another way to accomplish this would be to introduce attributes for these dimensions. For example, we could introduce the attributes Abundance-Of-Natural-Resource and Is-Exploited, and assign "small unexploited deposits of iron" the values Low-Abundancy and False correspondingly. Note that the two approaches can be made semanti-cally equivalent by introducing axioms such as:(<=> (low-abundance-natural-resource ?x)(abundance-of-natural-resource ?x low-abundance))3.2 Ontology development goals and criteriaTaking into account that taxonomies of these types of objects can have their own values, we decided to distinguish between two separate tasks: taxonomizing the terms in the WFB, and building a useful ontology of, say, natural re-sources. The goal was to do both. However, many of the terms in the WFB are not useful enough to be present in a general-purpose ontology. Some of them are direct intersections of other classes ("small deposits of coal"), and some terms are just too bizarre ("two small steel rolling mills" as an industrial sector of Saudi Arabia). To achieve both goals, the result of this stage of knowledge base building was subdivided into two smaller subontologies for all taxonomies constructed. The terms that were present in the WFB, but had no particular value as classes in a general-purpose ontology were exiled to one of the subontologies. Each of these terms is either an alias for a term in the main ontology, or a subclass that was considered too narrow to be retained (e.g. "Coastal-Climate-And-Soils-Allow-For-Important-Tea-And-Citrus-Growth" was abstracted to “fertile soil”).If the reasoner that is using the WFB has an effective means to deal with equalities, then the simplest solution is to assert such redundant terms equal to their more useful synonyms. However, with our reasoning tools we used simple rewrite rules: before importing the knowledge base into the theorem prover the facts file undergoes a separate pars-ing step when the unwanted synonyms are replaced with appropriate preferred terms,Some terms have a particular relation to terms in some different WFB taxonomy, for example the hydropower po-tential as a natural resource is used by energy industry, and the refined product of bauxites is aluminum. We tried to capture these connections.3.3 Criteria for introducing new termsSometimes several classes present in the WFB have an evident common superclass that fits nicely in the general taxonomy, but does not occur explicitly in the original document. In these cases, we introduced the missing class. An example is fossil fuels as a class of natural resources, with subclasses: petroleum, coal, natural gas and shale oil.4 Ontology ConstructionThe base representation used by the WFB, and retained by the parser is not always ontologically well motivated, well suited for automated inference, or compatible with existing ontologies, such as the HPKB Upper Level ontol-ogy or the ontologies found in the Ontolingua ontology library [4]. For this reason, we shift the representation used in the WFB into one that is compatible with existing ontologies and is well suited for inference. It would be possi-ble to effect this shift within the parser itself. We chose, however, not to burden the parser with the task of refor-mulating the semantic representation. Instead, we use lifting axioms, which lift the facts from the base representa-tion into our target ontology. The target representation is strongly object oriented and includes definitions from the HPKB Upper Level ontology as well as from the Ontolingua ontology library. The use of lifting axioms to shift representations is common in knowledge rich approaches to information integration.Consider the following typical simple lifting axiom:(=>(and(wfb::area ?georef total-area (* ?n sq-km))(number ?n))5(has-total-area ?georef (* ?n square-kilometer)))There are three important properties of this axiom. First, it shifts from a relational model in which area is a three place predicate, to an object model. Second, the predicate vocabulary in the base facts is distinct from the predicate vocabulary used elsewhere in the knowledge base. Predicates used in the base facts are prefixed with “wfb::”. Third, the object vocabulary is actually retained during the representation shift. In this case the ?georef object will be present in both the base and target vocabularies. Finally, the axiom uses a representation of quantities and units from the Standard-Units ontology [8] from the Ontolingua library. In this ontology, mathematical operations, such as multiplication, are polymorphically extended to apply to units which have dimensionality, as well as numbers. Thus, (* 3 meter) is distinct from (* 3 liter), but identical to (* 0.003 kilometer). An additional function, magnitude, may be applied to a quantity and a unit to compute the numeric magnitude of that quantity given the unit.The following lifting axiom is somewhat more complex(=>(and(wfb::age-structure-by-sex ?georef 0-14-years female ?number)(number ?number)(population ?georef ?pop))(exists ?af(and(fraction-of-country-population ?af)(fraction-of ?af ?pop)(fraction-defining-set ?af age-0-14)(fraction-defining-set ?af female-person)(:cardinality fraction-defining-set ?af 2)(set-cardinality ?af ?number))))The WFB decomposes populations by both age and gender. This axiom lifts the base predicate wfb::age-structure-by-sex into the object oriented model used in the target ontology. In order to accomplish this, it must introduce a new object, ?af, that represents the subset of the population of a country. The fraction vocabulary introduced in Section 5.2 is used for this purpose.The classes and relations in the WFB ontology are hooked up to the HPKB Upper-Level Ontology. This is a fairly large ontology (on the order of 2000 classes and 800 relations). It is used as a common representation within the High Performance Knowledge Base project. Only a small part of the Upper Level ontology was used in the WFB KB (on the order of 20 terms out of more than 3000). This is not altogether unexpected, because the Upper Level ontology covers a much broader range than the WFB. Consequently, most of its classes and relations were irrele-vant for our purposes. The other ontology that we used was the KSL Standard-Units ontology for the representation of the physical quantities and units of measure [8].5 Representation issues5.1 TimeThe representation of time in the WFB KB is a topic deserving careful attention. The WFB itself is temporal in na-ture. A new release of the WFB is made each year. In addition to changes in format, one anticipates changes in the values of many of the fields. For example, the GDP, population, birth rate, and many of the quantitative measures change on an annual basis. Other facts may also change on a regular, but less frequent, basis. For example, the membership of the UN Security Council changes every two years (except for the six permanent members). Other facts change irregularly and with less frequency. For example, the flag of the United States changes when new states are added. Still others change with even less frequency. For example, Greece has bordered on the Mediterra-nean Sea, and Egypt has produced textiles since ancient times.There are several options for representing the temporal aspects of the WFB KB contents. The simplest is to assume that the facts in the KB hold over the year with which the WFB associated and to explicitly assert that each fact holds during that year. For example, every fact in the 1995 WFB might be asserted to hold throughout the year 1995. Even this simple approach admits to several possible realizations, such as (holds-in 1995 (capital-of France6Paris)) using a modal operator holds-in, or (capital-of 1995 France Paris) using an explicit temporal argument to each predicate. These temporal assertions, of course, require an accompanying set of axioms to relate the truth of a proposition over the year to its truth during the year. One could, for example, state that each predicate holds densely throughout the time interval:(=> (and (capital-of ?t1 ?x ?y) (subinterval-of ?t1 ?t2))(capital-of ?t2 ?x ?y))Although this sort of axiom holds for the relation capital-of, it does not, in general, hold for many of the predicates within the WFB.•Many predicates apply to an entire year as an aggregate. For example, the GDP of Japan during 1995 is not the same as its GDP during the month of July. Other measures, such as birth rate use mid-year estimates to com-pute a full-year measure, and are not meaningfully applied to any interval other than the full year.•Some values are not measured annually. For such predicates, the value listed in the WFB may actually be the value for a previous year. The old value is retained because it is the best available. For example, the budget fig-ures for Austria are stamped with year 1993.•Other values are computed based on estimates from previous years. GDP growth, for example, is the difference between the GDP for the current year and GDP for the previous year.Thus, it is simply not correct to associate each fact with the temporal interval associated with a particular edition of the WFB. It is more correct to observe that each fact is true in the context of a particular edition. We can make this notion of context explicit by using a context logic and asserting each fact within a context associated with a specific edition of the WFB. For example: (is-true (wfb 1995) (capital-of France Paris)). In the case that all of the facts are true in a single context and there are no assertions explicitly relating different contexts within the WFB, the context assertion can effectively be elided. This is the choice that has been made in Version 0.8 of the WFB KB. Another motivation for this choice is that for many problem-solving scenarios, the time interval under consideration is much smaller than a year. In such scenarios, it is desirable to treat the content of the WFB KB as fixed or atemporal. It is primarily for problem solving scenarios in which the temporal interval under consideration spans multiple years that the explicit representation of time is valuable.The next step in the WFB KB development, however, is to make the contextual and temporal properties of each predicate explicit. By combining the contextual information with properties at the level of predicates, we intend to support a reformulation process in which the WFB KB contents can be rewritten using either modal operators, ex-plicit temporal arguments and axioms, or atemporally as in Version 0.8.5.2 Sets and QuantitiesWhen we examine the information in the WFB, we observe that many of the data values measure properties of sets of objects associated with a country (e.g., the number of airports). Furthermore, many of these sets are partitioned into subsets. Sometimes there are several partitions according to different criteria. For example, the airports are par-titioned into paved and unpaved, as well as into three classes according to the runway length. Another example is a country’s exports, which are partitioned into classes based on commodities, as well as classes based on the partner country.Representing the relationships between objects, the sets associated with these objects, and the properties of these sets is critical in the WFB. Cardinality is one example of a property associated with some of these sets. We can identify a class of such measures, most of which are additive.The general case seems to be when there is some set associated with the country. The set has some additive quantity related to it, which we call the additive set function. In the general case it is a weighted sum over the set’s elements. For example, for population statistics, the weights are all equal to one and the additive set function is the cardinality of the set, i.e. the number of people in the country. For a country’s exports, the additive set function is the total dol-lar amount of sale. If the export is viewed as a set of business transactions, then the weight of an element will be the dollar amount of a particular transaction.7。

OWL Ontologies and SWRL Rules Applied to Energy Management Ana Rossell´o-Busquet,Lukasz J.Brewka,Jos´e Soler and Lars Dittmann,IEEE members Networks Technology&Service Platforms group,Department of Photonics Engineering, Technical University of Denmark,2800Kgs.Lyngby,Denmark{aros,ljbr,joss,ladit}@fotonik.dtu.dkAbstract—Energy consumption has increased considerably in the last years.How to reduce and make energy consumption more efficient in home environments has become of great interest for researchers.This could be achieved by introducing a Home Energy Management System(HEMS)into user residences.This system might allow the user to control the devices in the home network through an interface and apply energy management strategies to reduce and optimize their consumption.Further-more,the number of devices and appliances found in users resi-dences is increasing and these devices are usually manufactured by different producers and can use different communication technologies.This paper tackles this problem.A home gateway which integrates different devices with a HEMS based on set of semantic-web tools providing a common control interface and energy management strategies.Index Terms—Home gateway,Home energy management sys-tems,ontology,OSGi,engine rule,electrical grid,reasonerI.I NTRODUCTIONA home network is a residential local area network(LAN) containing all the devices in the home environment and may also provide connection to the internet or other networks. This network can be composed by different types of devices which provide a variety of functionalities and applications. In addition,these devices can be manufactured by different producers and can use different communication technologies. Therefore,the main challenge in home networks is the variety of technologies,providing different communication methods, as well as the diversity of producers,providing different types of devices and services.A Home Energy Management System(HEMS)[1]is a system from which the user can control the devices in the home network through an interface and apply energy man-agement strategies to reduce and optimize the devices energy consumption.The herein presented home gateway accesses a knowledge base data repository from where devices capabil-ities can be obtained.Furthermore,based on the knowledge implemented by an ontology,the home devices are classified according to their functionalities and capabilities.Moreover, the ontology also contains a classification of the different commands and status a home device can have.This offers interoperability between the different devices at the service level.Energy management strategies can be performed by applying a set of rules which are based on home devices consumption,information from the electrical grid and user’s preferences.In addition,the home gateway offers a graphical user interface(GUI)from where the user can interact with all the devices in the home network.The users can also define rules to apply energy management strategies to minimize the energy consumption of their home devices.In this article,each logical component contained in the the developed home gateway is described.The logical component containing the knowledge based data repository is explained in detailed.This paper gives implementation details of how the ontology is access and queried to obtain information about the devices contained in the home network.Furthermore,rules are applied to create the energy management strategies which makes this home gateway suitable for HEMS.The rest of this article is organized as follows:section II introduces relevant related works,while section III briefly de-scribes the OSGi framework.Section IV provides an overview of the home gateway architecture and section V gives a detailed description about the logical component containing the knowledge based data repository.Section VI describes the implementation of SQWRL queries and section VII gives details about the integration of SWRL into the system to apply energy management strategies.Finally section VIII provides final remarks and conclusions.II.R ELATED WORKThe implemented home gateway uses Open Service Gate-way Initiative(OSGi)[2]running over Equinox which is an Eclipse project that provides a certified implementation of the OSGi.In addition,OWL ontologies[3]have been used to create the knowledge base data repository for the implementa-tion by incorporating Prot´e g´e-OWL API3.4.4[4]mechanisms into our source code.OSGi and ontologies have been used previously to create home gateways.An example of this is Domotic OSGi Gateway(DOG)[5]by Politecnico di Torino. DogOnt[6]ontology is used to implement the knowledge base data repository.This ontology is reused in this implementation as it provides a good classification of the devices that can be found in a home environment.However,DOG is centered in domotics while the application of this implementation is focus on energy management.This is archived by offering the users the opportunity to create their own energy management system by creating,modifying and deleting rules which may reduce the total electricity consumption.Another example of a home gateway developed using OSGi can be found in[7].However,ontologies nor rules are not used there,therefore energy management strategies cannot be applied.The authors in[8]describe the IntelliDomos learning model which is an ontology-based system able to control a home automation system and to learn users periodic patterns when using home appliances.However,the authors do not describe2011 UKSim 13th International Conference on Modelling and Simulationhow they deal with the fact that an ontology cannot be dynamically modified[9],which we will explain in more detail in section VII.III.OSG IThe OSGi Framework is an open service platform for the delivery and control of different Java-based applications, called bundles.Bundles are JARfiles containing extra meta-data which makes them fully self-describing.Bundles consist of packages,which containing Java classes,and a MAN-IFEST.MF,which contains the metadata.Packages are by default private,however,they can be used by other bundles if they are exported by the bundle containing them.Exporting a package makes it visible to other bundles,but theses other bun-dles must import the package in order to use it.Furthermore, OSGi framework allows bundles to register services so that other bundles can use them.A service is defined using a public Java interface which must reside in an exported package.A bundle then implements the service interface,instantiates it and then registers the instance using the OSGi Service Registry under the service interface name.The class that implements the service is usually private and not available to the other bundles.However,by using the Service Registry and importing the service interface,other bundles can import and consume the service.IV.H OME G ATEWAY A RCHITECTUREThe home gateway has been developed using OSGi-Equinox to control the devices of a home network and to apply energy management strategies.DogOnt ontology is used to create the initial knowledge base data repository of our home gateway. However,this article will not describe this ontology as an extensive description can be found in[6].Furthermore,the home gateway consists of the following bundles,which are shown in Fig.1:•Knowledge Base Bundle:This bundle handles all the interactions with the knowledge base data repository and rule engine.An ontology is used to implement the home gateway knowledge base data rmation about the devices,such as their capabilities,status and network parameters among others can be obtained from the knowledge base data repository by using queries.In addition this bundle contains a rule engine which executes rules in order to apply the energy management strategy chosen.•Interface Bundle:This bundle is used to create a graphical user interface(GUI),from where the users can communicate to the devices and obtain information about them.Moreover,users with advanced knowledge about the underlying ontology can introduce more devices into the knowledge base data repository and modify the rules in the system.•Network n Bundle:As mentioned before more than one communication technology,for example power line or wireless,maybe be found in the home network.Each of these n bundles will handle the communication with theFig.1.Home Gateway Architecturedevices in the subnetwork n in the home network.These bundles will send messages to the devices and forward notification messages from the devices,behaving as tech-nology bridges.For example,a notification message could contain information about a device changing status.These bundles have not been implemented in the work herein described as the aim of this paper is to test the bundles design and the interaction with the ontology.However,a bundle to support communication with KNX networks is under construction.•Networks Manager Bundle:This bundle handles the communication with the different Network bundles.When a message has to be send to a device,this bundle will receive a notification and forward the message to Network bundle containing the recipient device.•Manager Bundle:This is the central bundle which handles the interaction between the different bundles that make up the home gateway implemented.•Network Emulator Bundle:As currently no devices are connected to the home gateway,this bundle emulates the network and its devices and will allow us to test the home gateway implementation and its specific application. More information about the home gateway architecture can be found in[10],where the interaction between bundles and each bundle is explained in more detail.V.K NOWLEDGE B ASE B UNDLEThe Knowledge Base bundle is the bundle in charge of the knowledge base data repository and rule engine.It manages the interaction between the knowledge base data repository and the rule engine with the rest of the architecture.An ontology has been used as knowledge base data repos-itory.This ontology contains the information of the different devices that can be found in the home network,including the devices features and functionalities.The ontology used is DogOnt.For each device found in the home network an OWLIndividual is created.This OWLIndividual is an instance of the corresponding class in the ontology,which represent the actual device found in the home network.Furthermore, OWLIndividuals representing the device’s functionalities and status are also created and linked by object properties to the OWLIndividual of the device as seen in rmationabout devices is accessed by using SQWRL(Semantic Query-Enhanced Web Rule Language)[11],which query the knowl-edge base data repository to obtain the necessary information. Users are not able to access or modify the ontology or the queries directly.However,a user can obtain information about the devices and their capabilities and introduce new OWLIn-dividuals into the knowledge base data repository through the GUI.In addition,in this home gateway,rules can be defined to reduce the energy consumption of the devices creating therefore a HEMS.A rule will be triggered when certain conditions are satisfied and will apply the energy saving strategy defined.Jess Rule Engine[12]is used to execute rules.The rules are written in SWRL(Semantic Web Rule Language)[13].The user is able to create,modify and delete SWRL rules in the system through the GUI.In order for users to create valid rules,they need to have some knowledge about the ontology and SWRL language.For accessing the OWLfiles and creating the knowledge base data repository Prot´e g´e-OWL API3.4.4[4]is used.This API provides methods to load the ontology from an OWLfile, modify the ontology and get classes and OWLIndividuals.In addition,Prot´e g´e-OWL API allows to run SQWRL and define SWRL rules that are executed by using Jess Rule Engine.This API is integrated into this bundle to enable all this capabilities.A.Knowledge Base Operator ServiceThe Knowledge Base Operator Service is offered by the Knowledge Base bundle.This bundle exports the package containing the Java interface of this service.Moreover,it implements the service interface,instantiates it and then reg-isters the instance using the OSGi Service Registry under the service interface name.The Knowledge Base Operator service offers access to the knowledge base data repository allowing the subscribed bundles to query and obtain information about the devices and energy management rules.This service is only used by the Manager bundle as a mid-step to query the knowledge base data repository.This separation between the Knowledge Base Operator service,which directly accesses and queries the knowledge base data repository,and the Manager bundle,used by bundles that need to obtain information contained is done to facilitate the modularity of the system. This is done to avoid interoperability problems if the Knowledge Base bundle is modified.Hence,a separation between the Knowledge Base Operator service,which directly accesses and queries the knowledge base data repository,and the Manager Functions service,used by bundles that need to obtain information contained in the knowledge base data repository is made.VI.Q UERYING THE K NOWLEDGE B ASE D ATAR EPOSITORYProt´e g´e-OWL API provides also methods to load,create and execute SQWRL queries,which are used to obtain information from the knowledge base data repository.SQWRL is based onFig.2.System ArchitectureSWRL language for querying OWL ontologies and provides SQL-like operations to retrieve knowledge from them.The SQWRLQueryAPI[14]is used to run and retrieve the result of SQWRL queries.This API is available in the Prot´e g´e-OWL API.However,in order to run SQWRL queries,Jess rule engine is needed.Jess is not open source and must be downloaded separately.Furthermore,Jess rule engine requires a license,which is free for academic use.Fig.2shows,in section a,part of DogOnt ontology and a partial sample lamp model in DogOnt in section b.The Controllable class is the class representing devices or home appliances that can be controlled.Any OWLIndividual of the class Controllable is linked to another OWLIndividual of the class Functionality with the object property hasFunctionality. The Functionality class is used to represent the functionalities of the device.Furthermore,to select between the different functionalities of a device a command is needed,which is represented with the Command class.Functionality OWLin-dividuals are linked to Command OWLIndividuals with the object property hasCommand.RealCommandName is a data property which links the Command OWLIndividuals to a string which is a human readable string.In addition,Control-lable OWLIndividuals are also linked to State OWLIndivid-uals by hasState and to BuildingEnvironment by isIn.The State class is used to represent the state of the device and contains a data property,value,which represents the actual state of the device.The BuildingEnvironment class is the representation of domestic environments,such as rooms in the user premises.For instance,consider that the knowledge base data repos-itory contains a device Lamp Bedroom,which is located in the bedroom and has only On/Off functionality.To obtain the state in which the Lamp Bedroom is at a certain point in time without sending a command to it the following SQWRL query can be used(to simplify DogOnt namespaces have been omitted):hasState(Lamp Bedroom,?Lstate)∧value(?Lstate,?value)→sqwrl:select(?value)This query searches the knowledge base data repository to find the State OWLIndividual linked to the Lamp Bedroom device and returns the string stored in the data property value.SQWRL queries are also use to retrieve some other infor-mation from the knowledge base data repository.For instance, the following query is used to retrieve the valid commands that can be send to a the device Lamp Bedroom.hasFunctionality(Lamp Bedroom,?Lfunc)∧hasCommand(?Lfunc,?Lcmd)∧realCommandName(?Lcmd,?LcmdName)→sqwrl:select(?LcmdName)This query will return two results one will return the value On and the result will return the value Off.Depending on how many commands the device has,this query will have different number of results.Furthermore,by changing Lamp Bedroom for?device the query will return all the possible commands for all the devices found in the knowledge base data repository.This query is used by the GUI when it is initialized as all the commands are displayed in a combo box.VII.R ULE E NGINESWRL(Semantic Web Rule Language)is based on a combination of the OWL-DL and OWL-Lite,which are sub-languages of the OWL.SWRL can be used to reason about OWLIndividuals by using OWLClasses and properties terms. SWRL is therfore used to create rules which will be used to achieve energy management strategies.To implement energy management strategies a rule engine is used.The user can create,modify and delete rules through the GUI.SWRLRuleEngineAPI[15],[16]is provided in Prot´e g´e-OWL API for creating and manipulating SWRL rule engines.Prot´e g´e-OWL API supports inference with SWRL rules using the Jess Rule Engine.SWRLJessBridge[17]is also provided in Prot´e g´e-OWL API to allow the interaction between the knowledge base data repository,SWRL rules and Jess rule engine.SWRLJessBridge is used to perform syntax transformation from OWL syntax to Jess syntax.Furthermore, SWRLJessBridge calls Jess Rule Engine,which performs Jess reasoning and returns assertion results to SWRL Bridge. Even though SWRLJessBridge is provided in Prot´e g´e-OWL API,Jess rule engine must be downloaded separately and an academic license must be obtained in order for it to work. SWRL rules have a drawback for this home gateway imple-mentation.The state of the devices is saved in the knowledge base data repository by using OWLIndividuals of the State class.As stated in[9],”SWRL rules cannot be used to modify existing information in an ontology”.For example,assume we have a rule that indicates that if the light intensity in a room is higher than50%of the capacity of the sensor,the lamp in that room should be turned off(to simplify the DogOnt namespaces have been omitted):BuildingEnvironment(?room)∧Lamp(?lamp)∧isIn(?room, ?lamp)∧hasState(?lamp,?Lstate)∧value(?Lstate,?value)∧swrlb:stringEqualsIgnoreCase(?value,”On”)∧LightSensor(?sensor)∧isIn(?room,?sensor)∧hasState(?sensor,?Sstate)∧value(?Sstate,?intens)∧swrlb:greaterThan(?intens,50)→value(?Hstate,”Off”)Fig.3.Rules WindowThis rule will add the value of”Off”to the value property for lamp state.This rule will not change the existing value for that property,so a successfulfiring of this rule will result in the property value having two values.In order to solve this problem SWRLBuiltInBridge[18],contained in Prot´e g´e-OWL API,is used.SWRLBuiltInBridge provides a mechanism that allows the use of user-defined methods in rules.These methods are called built-ins and can be seen as functions of the rule system.These user-built-ins are implemented in a java class and dynamically loaded into the rule engine,which can then invoke them when necessary.In order to create user-defined built-ins,first an ontology containing the definition of these user-built-ins must be created and imported into the java implementation.The package name should end with the name of the ontology defining the built-ins.Additionally,the SWRL-BuiltInLibraryImpl class must extend the abstract base class AbstractSWRLBuiltInLibrary.The user-built-ins definition for our implementation are contained in owlmod.owl and imple-mented in the java SWRLBuiltInLibraryImpl class contained in edu.stanford.smi.protegex.owl.swrl.bridge.builtins.owlMod which must be included in the Prot´e g´e-OWL API.These user-built-ins can be used in SWRL rules by calling them using owlmod:prefix and then the name of the method found in the java class.This class will also import the knowledge base data repository in order to be able to change the information about the devices.In order for the above rule to work dogont:value(?Lstate,”Off”)is replace by owlmod:changestate(?Lstate,”Off”)All rules will be evaluated every time the knowledge base data repository is changed.For instance,if the home gateway receives a notification of a sensor sending a status update,Jess rule engine will evaluate and trigger the necessary rules. The GUI offers the user the possibility to view the system rules,create new rules,modify them,or delete them at any moment.This window is shown in Fig.3.Furthermore,when the user clicks on create or modify a new window opens where the rule chosen,in case of pressing modify,will be displayed as shown in Fig.4.In case ofFig.4.Modify Windowpressing create the window will be empty and the user can introduce the new rule.If the rule the user introduces is invalid an error message will be display giving the user details about the error in the rule.The method of creating energy management strategies by using rules gives users the freedom to define different rules adapting the system to the users requirements and personal preferences.VIII.C ONCLUSIONIn this article we have presented the home gateway devel-oped using OSGI-Equinox framework and Prot´e g´e-OWL API. An overview of the bundles composing the home gateway has been given.Furthermore,SQWRL has been used to create queries to extract information from the knowledge base data repository.On the other hand,SWRL is used for rules which can be created,deleted and modified to apply en-ergy management strategies according to the users’priorities. SWRLBuiltInBridge has been used to create our own built-ins to enable rules to modify the knowledge base data repository when necessary.This home gateway provides the means to create energy management strategies which makes this home gateway suitable for HEMS.R EFERENCES[1] A.Rossell´o-Busquet,Georgios,J.Soler,and L.Dittmann,“TowardsEfficient Energy Management:Defining HEMS,AMI and Smart Grid objectives,”in The Tenth International Conference on Networks,Jan.2011.[2]OSGi Alliance,“OSGi Service Platform Core Specification Release4,”Accessed Dec.2010.[Online].Available:[3]W3C OWL Working Group,“OWL2Web Ontology LanguageDocument Overview,”Accessed Dec.2010.[Online].Available: /TR/owl2-overview/[4]H.Knublauch,“Protege-OWL API Programmer’s Guide,”AccessedDec.2010.[Online].Available:/wiki/ProtegeOWL API Programmers Guide[5] D.Bonino,E.Castellina,and F.Corno,“The DOG gateway:enablingontology-based intelligent domotic environments,”Consumer Electron-ics,IEEE Transactions on,november2008.[6] D.Bonino and F.Corno,“DogOnt-Ontology Modeling for IntelligentDomotic Environments,”The Semantic Web-ISWC2008,2008. [7]S.Ok and H.Park,“Implementation of initial provisioning functionfor home gateway based on open service gateway initiative platform,”in Advanced Communication Technology,2006.ICACT2006.The8th International Conference,feb.2006.[8]P.Valiente-Rocha and A.Lozano-Tello,“Ontology and SWRL-BasedLearning Model for Home Automation Controlling,”in Ambient Intel-ligence and Future Trends-International Symposium on Ambient Intel-ligence(ISAmI),ser.Advances in Soft Computing.Springer Berlin/ Heidelberg,2010.[9]M.J.O’Connor,“SWRL LanguageFAQ,”Accessed Dec.2010.[Online].Available:/cgi-bin/wiki.pl?SWRLLanguageFAQ [10] A.Rossell´o-Busquet,J.Soler,and L.Dittmann,“A Novel HomeEnergy Management System Architecture,”UKSim13th International Conference on Computer Modelling and Simulation,2011.[11]M.J.O’Connor,“SQWRL,”Accessed Dec.2010.[Online].Available:/cgi-bin/wiki.pl?SQWRL[12]Sandia National Laboratories,“Jess Rule Engine,”Accessed Dec.2010.[Online].Available:/[13]I.Horrocks,P.F.Patel-Schneider,H.Boley,S.Tabet,B.Grosof,andM.Dean,“S.”[14]M.J.O’Connor,“SQWRLQueryAPI,”Accessed Dec.2010.[Online].Available:/cgi-bin/wiki.pl?SQWRLQueryAPI [15]——,“SWRLRuleEngineAPI,”Accessed Dec.2010.[Online].Available:/cgi-bin/wiki.pl?SWRLRuleEngineAPI [16]——,“SWRLRuleEngineBridgeFAQ,”Accessed Dec.2010.[Online].Available:/cgi-bin/wiki.pl?SWRLRuleEngineBridgeFAQ[17]——,“SWRLJessBridge,”Accessed Dec.2010.[Online].Available:/cgi-bin/wiki.pl?SWRLJessBridge [18]——,“SWRLBuiltInBridge,”Accessed Dec.2010.[Online].Available:/cgi-bin/wiki.pl?SWRLBuiltInBridge。