Modeling User Context with Applications to Media Retrieval
Ankur Mani Hari Sundaram
Arts Media and Engineering Program
Arizona State University
e-mail: {ankur.mani, hari.sundaram}@https://www.doczj.com/doc/9c7079709.html,
Abstract
In this paper, we develop a theoretical understanding of multi-sensory knowledge and user context and their inter-relationships. This is used to develop a generic representation framework for multi-sensory knowledge and context. A representation framework for context can have a significant impact on media applications that dynamically adapt to user needs.
There are three key contributions of this work: (a) theoretical analysis, (b) representation framework and (c) experimental validation. Knowledge is understood to be a dynamic set of multi-sensory facts with three key properties – multi-sensory, emergent and dynamic. Context is the dynamic subset of knowledge that affects the communication between entities. We develop a graph based, multi-relational representation framework for knowledge, and model its temporal using a linear dynamical system. Our approach results in a stable and convergent system. We applied our representation framework to a image retrieval system with a large collection of photographs from everyday events. Our experimental validation with against two reference algorithms indicates that our context based approach provides significant gains in real-world usage scenarios. Keywords
User context, media retrieval, multi-sensory knowledge representation
1.INTRODUCTION
In this paper we develop a representation framework for multi-sensory knowledge and user context, and apply this framework in a media retrieval application. The development of a generic representation for context is an important problem in multimedia. Intuitively, context is critical in explaining the “why” in communication – i.e. it represents the set of concepts that fully explain the concept that is contained in the message. Thus context play an important role in the exchange of messages in any communication or interaction. In face-to-face human communication, contextual cues play a key role [13,31], allowing us to act and participate in conversations in a rich manner. Similarly, a flexible representation framework that additionally addresses temporal evolution of context, can lead to powerful media applications that gracefully adapt to user interaction.
In this paper, we focus on developing a theoretical understanding of context, and its relationship to multi-sensory knowledge. Then we develop a graph based representation of knowledge, and show how we can model context as a dynamic subset of knowledge. We apply this framework on a photo browsing application with a large database of photographs from ordinary everyday events. We now summarize the unique and novel contributions of this paper are as follows.
Theoretical analysis: We present a detailed understanding of multi-sensory knowledge and
context in the communication between two
entities. We define knowledge as a dynamic set of
multi-sensory facts. There are three key aspects of
knowledge – (a) multi-sensory, (b) emergent and
(c) dynamic. We define context as “the finite and
dynamic set of multi-sensory and inter-related
conditions that influences the exchange of
messages between two entities in communication.”
Context forms a dynamic subset of multi-sensory
knowledge that is central to the semantics of the
communication.
Representation framework: We develop a graph based representation of multi-sensory knowledge.
Each concept is a node in this graph, and a pair of
nodes can have multiple relationships (feature
based as well as semantic and each relationship
represented as a weighted edge). Context is
represented as the subset of nodes in attention that
affect the communication. A key innovation is the
development of a linear dynamical system to
model the evolution of knowledge. We show that
the system is stable and convergent.
Experimental validation: We have applied our representational framework to a practical image
retrieval application. The dataset comprised a
large collection (~4000) of personal photographs
of ordinary everyday events. This dataset is
challenging as it is not organized into discrete
categories, but is a better indicator of real-world
results. We tested our algorithm of context aware
retrieval against relevance feedback [27] and
random browsing (baseline). Our user studies with
graduate students indicate that our approach has
significant gains.
It is important to note that the specific attributes of context will always be application dependent, driven by the real-world application needs.Our research instead focuses on a generic representation framework that allows for a systematic mechanism to model knowledge with arbitrary
attributes and their temporal evolution. This allows our contributions to context and knowledge be applied to other applications.
The rest of this paper is organized as follows. In the next section we begin by developing our notion of context – we provide real-world examples and then define context, multi-sensory knowledge and their inter-relationships. In section 3, we discuss related work. In section 4, we develop the key contribution of this paper in terms of a context model for media retrieval. In section 5, we present the architecture of our photo retrieval application. In section 6, we present our experiments on a large real-world data set and in section 7, we discuss the limitations of this work. Finally in section 8, we present our conclusions and future work.
2. MULTI-SENSORY CONTEXT
In this section we introduce the definition of multi-sensory and discuss some of its properties. We first provide two examples that demonstrate the context and the role of context. Then, we define context, multi-sensory knowledge and its relationship to context.
2.1 Examples of Context
We now discuss two user tasks and determine context in each task and then discuss the role that context plays in fulfilling the tasks.
Media retrieval: In the first task, the user wants to search photographs in a media database containing photographs, music, movies and others. The key problem here is to deliver to the user a set of media of elements most relevant to the user. The user provides query as either text or other by selecting media through an interface that allows for both modalities.
There are concepts and events that help in the query formulation. The user has some information about the set
of events where such photographs were taken or the people or things visible in the photograph or how the photograph may look like. This set of information may provide a vague query, however most times it is difficult to provide query this way. Thus, the query is related to the current set of information in the user’s short-term memory [4]. The user associates the photographs to other available media in a unique way. The strengths of these associations change with time and are related to the way the user associates things in his short-term memory. The multi-sensory and interrelated information set in the user’s short-term memory influences the query provided by the user [4] and at the same time is influenced by the user’s activity and the media the user consumes. This set of multi-sensory and interrelated information forms the user query context . The system can use estimates of the query context to provide a more comprehensive set of results as a response to the query. In this example attributes that do not form the query context include the temperature of the room, the sound of the air conditioner, or the last news article read.
Human Communication: In the second example, let us examine the case of two people Mary and Jane engaged in a face to face communication about a legal contract. In this case, Mary’s spoken communication is highly dependent on several factors. It depends on the interaction immediately prior to Mary’s spoken utterance, as well as multi-sensory cues such as Jane’s gestures, clearly discernable affect states (e.g. anger). Mary incorporates estimates of Jane’s interpretation context when composing her next spoken utterance – Mary’s context of construction , is thus influenced by Jane’s context of interpretation , as well as their past message history. Examples of attributes in this conversation that are not contextual include the color of the sky, the name of Mary’s brother, or the name of Jane’s graduate school – these attributes are part of the
overall knowledge but are not relevant to the conversation
Figure 1: The general framework of multi-sensory communication between two entities. The intent of communication is dependent on the context of construction while the decoded intent depends upon the context of interpretation. We can draw a communication surface around each entity, through which all communication to the entity must pass.
about the legal contract.
Generalization: These examples can be generalized into the communication problem between any two entities [16]. When one person sends messages to the other person using
a set of the available media, the messages are influenced by the person’s current short-term memory, the task at hand, the messages the person has sent so far, the messages the person has received and the person’s knowledge about the subject and the other person. This set of conditions is multi-sensory (represented in more than one medium) and inter-related and influence the messages originating from the communicator forms the context of the communicator. The interpretation of these messages by the other person (receiver) also depends upon a similar set of conditions for the receiver (ref. Figure 1). This set of conditions forms the context of the receiver. The more is the overlap between the contexts of the two people, the more effective is the communication. Note that the context of construction and the context of interpretation always represent a small part
of the overall knowledge available to both entities. The communication is also influenced by the environment knowledge and the current state of the environment. Social knowledge such as language and the social codes form a part of the environment knowledge and influence communication [16].
2.2What is multi-sensory Context?
We define context as “the finite and dynamic set of multi-sensory and inter-related conditions that influences the exchange of messages between two entities in communication.” This is the set of conditions that can be estimated by the receiver based upon the messages from the originator and the set of conditions that influence the messages originating from the transmitter of messages. We observe in the above examples that the conditions were inter-related, dynamic, emergent (i.e. lead to the formation
of new knowledge) and multi-sensory and influence the origin and interpretation of messages.
In the above examples the set of conditions was a subset of
a large number of possible inter-related conditions. The superset of these conditions is knowledge about the communicating entities and the environment. We now discuss the definition of knowledge and its relationship to the context.
2.3Knowledge and its properties
Knowledge is a dynamic set of multi-sensory facts. A fact
is a statement that holds true for an entity (communicators or environment). Knowledge has three important properties; it is multi-sensory, emergent and dynamic.
Multi-sensory: The environment we live in represents knowledge in multiple senses. The
color, sounds produced, the texture, the shape, the
structure and organization of different
components, all these represent knowledge about
the object. The functions of objects and their
interactions also represent knowledge. Similarly,
in our minds, we represent knowledge in multiple
modalities. When we talk of the concept “bird,”
we associate it with the visual representations of a
bird, the chirping sound of a bird, act of flying and
others.
Emergent: With increasing interaction with the environment, we begin to develop new
associations and gather facts previously not
known to us. For example, consider a child
walking on the beach for the very first time. The
child who has never been exposed to the sea
(visuals / sound / wind and their inter-
relationships), or the specific nature of the sand,
quickly begins to learn the properties of this
environment and use it to navigate the beach.
Dynamic: Through interaction with the environment, we refine and modify our
understanding of the existing relationships
between the different observed concepts. For
example Mary may associate sunshine with
happiness, in a region where there is very little
sunshine, but this relationship may weaken if she
moves to a country with abundant sunshine.
In a media retrieval scenario, knowledge can be divided into three overlapping sets namely user knowledge, environment knowledge and the application knowledge. The user knowledge consists of the set of interrelated facts about the user such as the user interests and the set of associations the user has. The application knowledge consists of the application code and the media database. The application knowledge also influences the user knowledge as the user learns new facts about the media while exploring the database. The environment knowledge consists of the physical and social environment the user is in and that influences the way user knowledge has developed and will change in future.
2.4Context and its relationship to knowledge Context is the dynamic subset of knowledge that is in attention and influences the exchange of messages between the entities in communication (ref. Figure 2).
When two entities are communicating, each entity maintains an estimate of the knowledge and context about the other entity and the environment. The accuracy of the estimate influences the effectiveness of the communication. For example, if one person knows that the other person is deaf, he/she will communicate with the other person using sign language. The importance of communication for a reliable estimate of context is valid even in the absence of a human in communication. For example, in a location based adaptive system such as a mobile phone that displays active map information, it is critical to have a continuous stream of information about the location of mobile phone provided by a Global Positioning System especially when the user is highly mobile. For simplicity of notation we refer to the
estimate of knowledge as ‘knowledge’ and the estimate of context as ‘context’ in the rest of the paper.
Not all the knowledge about the entities is needed for a limited set of message exchanges. In fact, only a limited
subset of knowledge that covers all the possible contexts for the specific communication is sufficient. For example, when a child studying in first grade communicates with a physicist about birds, the physicist will not refer to his understanding of ‘plasma physics’ to communicate with the child. Additionally, the child need not be informed about the physicist’s knowledge about ‘plasma physics’.
In a media retrieval application the messages from the user are the selection of a particular media element and explicit textual query. These messages are influenced by the user’s current attention and the way she associates different media elements. For the retrieval application the user knowledge is circumscribed by the media concepts, and the commonsensical relations connected to those concepts. An estimate of this knowledge by the application can make it more efficient. The environment knowledge can be limited to the semantic knowledge about the environment that influences the user interaction. This can be the linguistic, social, physical and commonsensical knowledge. The user always has some estimate of the knowledge and context of the application, such as how to construct the query, what to expect from the application. A user context-aware media retrieval application will also maintain an estimate of the user knowledge and query context. We next review related work.
3. RELATED WORK
Early work on context [10,11,12,28] treat context as pieces of information like location, identity, activity and time and has been successfully used for the purposes of configuration and adaptation is the areas of context aware ubiquitous computing. An attempt to integrate all such information was made in [9]. There the context was modeled for ubiquitous computing environments and hence was limited to the anticipated needs of the relevant ubiquitous computing applications. The context was defined as “any information that can be used to characterize the situation of an entity, where an entity is a person, place or object that is considered relevant to the interaction between a user and an application, including the user and application themselves.” This definition is very broad as pointed out in [34]. In [34], the author bounded the definition of context as the set of set of information that is relevant for the current communication. This definition is still broad and does not focus on two important properties of construct – (a) Context is a dynamic construct [14] and (b) Context is related to knowledge and cannot be discussed independent of it [26]. Our definition of context recognizes these important properties of context.
The generic model for context is not a well posed problem [17]. In [17] the authors pointed out the trade-off between the desire for abstraction (to make modular and tractable systems) and the desire for context-sensitivity. This also points towards an irony in making a generic context model (making a generic context model is actually motivated by this desire for abstraction) which is actually a drift away from the context sensitivity. Thus the attributes of context can only be decided on a per application basis. In this paper we focus on models of query context, with a model easily adapted to other applications. The needs of context in a media retrieval application are concerned with semantic inter-relationships between concepts, which can be arbitrary. Also the relationships that we wish to explore in multi-media are linguistic, statistical as well as common sense rules such as from the CyC database [2]. We address these issues in our work on context models.
The context model would thus consist of two structures:
A knowledge base (personal or shared ontology)
consisting of the concepts and the relationships between them
A temporally evolving context representation that
is in relation to this ontology.
Most recent attempts on knowledge representation and context modeling fall into one of the two categories namely the logical representation [7,21] and statistical representation [24,33] The logical representations are mostly driven by linguistic interpretations and the construction of meaning in language. On the other hand, the statistical knowledge representations are driven by pattern classification applications. For multimedia
applications, we need a multi-modal knowledge
Figure 2: The figure shows relationship between knowledge, context with user interaction over a period of time. We see that the knowledge grows over time, and that context forms a dynamic, highly evolving subset of the knowledge.
representation that integrates different approaches in single modalities. In [5], the authors proposed a multimedia knowledge framework involving arbitrary relationships between concepts. They suggested a learning framework in which the topology of the knowledge framework and the relationships between the concepts are statistically learned into a Bayesian network. This gives an ontology depicting purely arbitrary statistical relationships between the concepts but does not provide a meaningful interpretation and evolution of the knowledge framework. The other attempts to represent semantic knowledge include Wordnet [23], Cyc [16] and OpenMind Commonsense [20]. These knowledge representations focus on one or more relationships such as lexical relationships and commonsensical relationships.
Based upon such knowledge bases, several attempts have been made to represent and use context. This has been used
in web-base learning environments, e-commerce applications, media recommendation applications, intelligent agents and others. However, the knowledge frameworks do not evolve dynamically with the user interactions in the system, nor do these models consider the personalized and shared nature of the multi-modal knowledge of each individual.
A significant survey on the content-based image retrieval has been given in [30]. Prior works on adaptive media retrieval have used relevance feedback [8,27] for a better query estimate and refinement of distance metrics in the low-level feature space. Recent work on retrieval has focused on context-aware similarities [35], however the work focuses on learning the similarity metric using a kernel trick and the model of context is not proposed. These works on media retrieval consider that image relationships exist in low-level feature space. However, most of the times people associate images semantically and not using low-level features. Media retrieval that uses semantics for retrieval [3,19] uses general and static knowledge bases like ConceptNet [20]. The problem with this approach is that it does not model the differences in the similarities between the media elements seen by different people in different contexts. We build a media retrieval system based upon the dynamic multimodal model of the user context as a subset of the personalized user knowledge. Our experiments show that media retrieval using the models of user context as in our approach produces better results as compared to relevance feedback as in [27].
Some recent media retrieval approaches [15,25] use a significant amount of prior knowledge for image retrieval.
In [15], the authors use a pure metadata approach for image retrieval. The description of the images is in from the metadata of the images. Hierarchically, structured ontologies, obtained from the domain and from the media annotation define a set of properties of the images such as time, place and others. The query is also represented in a similar structure. Creation of such ontologies and in particular the annotation of all the images in the database is
a significant problem. Often the image database is sparsely annotated. Hence, such approaches should be merged with content-based approaches (). The search and recommendations are then performed by a set of logical rules (recommendation rules, hierarchy rules and mapping rules). In [25] the authors present an interesting task based approach to define the context of the user query and the context of the images and their annotations. The authors also have a dynamic notion of task context that captures the user’s task history. However, the approach is driven by annotation and relies on continuous annotation by the users and hence several tools are provided for the purpose of annotation in the application. Both the approaches [15,25] need significant amount of information from the user continuously. On the other hand, our approach makes limited demands from the user and grows the knowledge automatically by analyzing the user interactions. This is very crucial for keeping the users interested and at the same time provides opportunities for integration with incentive based strategies [29]for gathering information, for knowledge growth, from the user.
4.CONTEXT FOR MEDIA RETRIEVAL
In this section we introduce our model for representation and evolution of multi-sensory knowledge (text, images) and user query context.
4.1Knowledge Representation
User knowledge in the model is represented as a graph as shown in Figure 3. The nodes in the graph are the instances
of concepts in one modality and the weighted edges (weights represent the similarity between the end nodes along the edge) are the relationships between those instances. The knowledge can be divided into planes as shown in Figure 3, where each plane consists of concepts
Figure 3: Multimodal User knowledge, nodes represent
concepts and edges the relationships between concepts.
in one modality and the relevant relations between them. The planes are connected through inter-planar edges between some nodes like an image in the images plane and
its annotation in the text plane.
The relations between concepts in the text plane are linguistic and commonsensical, in the images and audio planes are low-level feature based, in the people plane are social and in the location and time planes are spatio-temporal. For now the knowledge is restricted to only text and image planes.
The description of the knowledge graph is as follows. Let
C be the set of concepts in the knowledge and R be the set
of relation types among the concepts in the knowledge. Each concept c i∈ C and each edge e j∈ E = C x C x R. Each edge e j also has a weight s j(c m,c n) that denotes the similarity between the concepts c m and c n across the edge e j due to the edge e j. The weights of the edges in the text plane and the inter-planar edges are all set to 1 while the weights of the edges in the image plane are equal to the low-level feature based similarity between the images. The image plane is fully connected but the text plane is sparsely connected. The user knowledge is then defined as K = {C, E}.
Media knowledge is represented as a graph similar to the user knowledge and consists of the media elements and user specified annotations as the nodes and the feature based relationships (e.g. color histograms), commonsensical relationships and user specified annotation based relationships as edges. Environment knowledge at present is the commonsensical knowledge obtained from the ConceptNet [16] and it consists of text nodes and commonsensical relationships between those nodes.
4.2User Query Context Representation
User query context is represented as the subset of the nodes and edges in the knowledge graph that are in attention. Although the user knowledge consists of a large set of concepts and relationships, at any time only a few of these concepts are in the user attention with different levels of attention. The concepts influence the user’s interactions and the amount of influence depends upon their attention level. The association between the concepts is also different at different times and depends upon the level of attention on the relationships types connecting the concepts. An estimate of this set of concepts and relationships and their attention levels is represented as the user query context.
The attention on concepts and relationship types are represented as weights of the respective concepts and the biases on the types of relationships. The user attention at a concept c i in the knowledge is modeled as the weight w i of the concept. The sum of weights of all concepts in the user knowledge at any given time is constant. This is based upon the assumption that the amount of short-term memory
in any individual and hence the total amount of attention given to all the concepts is constant over a small duration.
The importance of a type of relationship r k∈ R is modeled
as the bias wr k on the relationship type. The biases on the relationship types at any point are estimated as
2(1)
k i j
i j
wr w w
αα
=+?∑∑, <1>
where w i and w j are the weights of the neighboring concepts c i and c j connected by the relationship r k and α is
a constant between 0 and 1. Thus wr k lies between 0 and 1.
Thus, a relationship type connecting concepts with high attention has a higher bias as compared to the relationship
type connecting concepts with low attention levels.
The similarity between the neighboring nodes is the weighted sum (weights are the biases on the types of edges) of similarities between the nodes along all the edges
11
1
(,)(,|)
N
i i j i i
j
S c c wr S c c j
++
=
=∑. <2>
S(c i,c i+1) is the similarity between the concepts c i and c i+1,
N is the number of edges between the two concepts and
S(c i,c i+1 | j) is the similarity of the j th relationship (along the
j th edge) between any two nodes.
4.3Context and Knowledge Initialization
Initial multimodal user knowledge is created using a set of multimodal (text, images and possibly other media such as
audio) seed concepts provided by the user and the relation between the concepts either provided by the user or already
present in the environment knowledge space. Each user provides some seed concepts describing herself such as her profession, interests, some images of herself and her friends, place and others. A set of concepts are extracted
from the environment knowledge space (ConceptNet
[1,20]) that are in the neighborhood of the textual seed concepts. The relationships between these concepts are also extracted from the environment knowledge space. All the
seed concepts and extracted concepts and the relationships between them form the initial User Knowledge. Initially the
whole user knowledge is set as the user context with equal
weight of the nodes and the edges. This is a conservative estimate of the user context as the system does not have
any information about the user’s current context. Thus initially, each concept in the user query context has equal weight.
4.4Context Evolution
We now discuss the evolution of the user context as the
change in the weights of the nodes and biases of the relationships and discuss the convergence and stability of
the model and intuitively discuss the appropriateness of the model. Earlier work on determining context of concepts in ontology has used activation spreading [7,8]. Thus the
context in these cases is static as opposed to real situations
where the context changes with time. We introduce a method for modeling the dynamics of activation spreading
that helps us determine and track context through time both
in the presence and absence of any information about the user activity.
The change in weights of the nodes and edges follow the change in attention on the concepts and relationships in the short-term memory and are modeled as a linear dynamical system. The weight of a concept in the knowledge is affected by the weight of its neighbors. A concept strongly connected with other concepts with higher attention, gains attention while a concept strongly connected with concepts with lower attention, loses attention. The user activity related to a particular concept increases attention at the concept at the cost of the attention over all other concepts in the knowledge. Thus the rate of change of weight of a concept is
1
((,))()()N
i i j j i i j dw C S c c w w A i Bw ==?+?∑ <3> where w i is the weight of the concept i at time t , S (c i ,c j ) is
the similarity between the neighboring concepts as given by equation <2>, c i and c j and C (x ) is a bounded monotonically increasing function. C (S (c i ,c j )) represents the coupling between two neighboring concepts that causes the flow of weights between the two concepts. A (i ) is the activation function that is non-zero only for a certain set of concepts related to the current user activity and B is a constant that is related to A (i ) as
1
()N i A i B ==∑. <4>
A (i ) and
B model the attention that is diverted from the existing set of concepts to the new set of concepts representing the current user activity. The first term of equation <3> models the spreading of the attention from the points of attention across the knowledge space. The context converges asymptotically on a suitable distribution of weights over all the nodes and edges in the knowledge under the presence of the messages from the user. We now discuss the steady state distribution of weights of
the concepts. The sum of weights of all the concepts is
constant. It can be verified using the equation <3> that
1
0N
i
i dw dt ==∑. <5> The state-space equations in <3> can be represented in the
vector form as (()))((()))d diag B dt
diag B =??+=?++w
Cw C1w w A C C1I w A , <6> where w is the vector of the concept weights, C is a matrix whose elements c i,j represent the coupling between the neighboring concepts c i and c j given as C (S (c i ,c j )), 1 represents a unit column vector and diag(x ) represents a diagonal matrix whose diagonal elements are the elements of vector x . A is the activation vector whose elements are A (i ). When no information about the user activity is
available, A is a zero vector and B is 0. Thus equation <6> reduces to
(())d diag dt
=?w
C C1w . <7> C is a sparse and symmetric matrix and the non-zero elements correspond to the coupling between neighboring concepts. At steady-state d w /dt = 0. Thus the steady-state solution lies in the right null space of the matrix C - diag(C1). The given matrix is the weighted adjacency matrix of the knowledge graph where the weights are the coupling values between the adjacent concepts. It can be easily verified that the rank of such a matrix C - diag(C1) is N R – N C : number of rows – number of connected components. Hence the dimension of the solution space (i.e. the null space of the matrix C - diag(C1)) is N C . The steady state solution thus is the uniform weight distribution across all the concepts in any connected component. In the current representation of knowledge, there is only one connected component in the User Knowledge network. Hence under the steady state, all concepts have the same weight. Further, since the sum of the weights of all the concepts is constant, this weight is unique and is 1/Number of concepts. This is intuitive since if we do not have any information about the user activity for a long period of time, then the estimate of the context is the just the overall knowledge that we have about the user.
4.4.1 Analysis of Stability and Convergence We now analyze the stability and convergence of the user
context in the general case when the information about the user activity is available and A and B is not 0. A Linear Time Invariant system with the state space equation as in eq. <6> is stable if all the eigenvalues of matrix C - diag(C1) – B I are less than or equal to 0. The system is not time invariant because the coupling matrix C changes with changes to context. However the change in matrix is small enough to be neglected and a time-invariant approximation
is still justified. It can be easily verified that diag(C1) – C is positive semi-definite given the elements of C are all positive. Adding a diagonal matrix to this matrix whose diagonal elements are all positive or 0 gives a matrix which is also positive semi-definite. Therefore, diag(C1) + B I – C is a positive semi-definite matrix with all eigenvalues positive or 0. Therefore, all eigenvalues of C - diag(C1) – B I are less than or equal to 0 and the system is stable.
In the general case when A is not 0, in the steady state, the weights of the concepts are the solution to the equation: ((()))diag B ?+=?C C1I w A . <8>
To determine if the dynamical system converges and an unique solution exists, let us consider the rank of C - diag(C1) – B I .
(())()T T
diag B V V B V B V ??=Σ?=Σ?C C1I
I I ,<9> where Σ is the diagonal matrix whose diagonal elements are the eigenvalues of (C - diag(C1)) and V is the matrix whose columns are the eigenvectors of (C - diag(C1)). We have discussed earlier that the rank of (C - diag(C1)) is N R – 1 as the knowledge graph has only one connected component. Thus, one of the eigenvalues is 0. From the discussion on stability, we also found that the eigenvalues of (C - diag(C1)) are less than or equal to 0. Therefore, all the eigenvalues of (C - diag(C1) – B I ) are negative (B is a constant non-negative number) and the matrix has full rank. Hence, the estimate of the user context converges on a unique distribution of the weights over the concepts. The context estimate peaks at a small subset of the knowledge closer to the concepts related to the information about the user activity. The sharpness of the peaks can be controlled by the value of activation vector.
An important property of our context model is that with suitable activation vector, any weight distribution across the concepts in the knowledge can be achieved. This is true because the dynamical system is controllable. To verify controllability, we use the Popov-Belevitch-Hautus test [6] which suggests that the dynamical system in eq. <6> is controllable if there exists no left eigenvector of (C - diag(C1) – B I ) that lies in the left null space of I . Since, there exists no null vector for I , therefore the system is controllable. In this section we discussed the evolution of the user context and analyzed the dynamics of the context. We now discuss the evolution of user knowledge.
4.5 User Knowledge Evolution
New facts about the user are exposed to the system through
interaction and at the same time the user is exposed to new media elements. Thus the user knowledge is an increasing set of concepts.
The user knowledge estimate changes as new facts about the user are discovered. Each activity of the user (for example a query) provides information about the user. If the query concept is not present in the user knowledge, the user knowledge is expanded as follows:
If the query concepts are textual, use them as seed
concepts, extract a set of concepts in the environment knowledge that are in the neighborhood of the query concepts. If the query concepts are images and have annotations, the annotations of the images or the annotations of the images form the seed concepts. If the images do not have annotations, add the images to the User Knowledge, no textual concepts or relationships are added.
Obtain the relations between the new concepts and
the existing concepts in the user knowledge. For
textual annotations use the ConceptNet neighborhood and add the concept and the relationships. For the image features, calculate the similarities with other images that are part of the user knowledge.
Add the new concepts and relationships to the user
knowledge.
The user knowledge estimate is also expanded as the user is exposed to new media elements. The new media elements that the user encounters and their annotations are added to the user knowledge. The new concepts are linked to the rest of the knowledge through the feature level relationships between the new media elements and the older ones in the user knowledge.
4.6 Similarity of non-neighboring concepts
The similarity of two non-neighboring concepts in the knowledge graph is computed as discussed in this section. Note that the similarity between the concepts depends upon the current context.
In the first case, when two concepts are in the current context, the similarity between them is computed as the maximum similarity between the two concepts computed over all the paths between the two concepts. Let p (k,m ) be a path that begins with concept c k and ends with concept c m . Let l (p ) be the length of the path. Then,
1
11(,)
1(,)max (,)(,),n k m n m i i i p k m i S c c S c c w S c c ???=??
=????
∏ <10>
where S (c i-1,c i ) is the similarity between the two successive concepts along the path and w i is the weight of the concept c i along the path and where n = l (p)+1, and c (0) = c k . This maximization can be done using a variant of the Dijkstra's shortest path algorithm. Thus, the path with the high attention concepts and strong similarity between the concepts is the maximum similarity path.
In the second case, when one or both concepts are not present in the current context we enlarge the context to encompass both concepts. The context is enlarged by the evolution process as given by equation <7> without any activation. During the evolution process all the concepts would lose their weights by a certain amount and the similarity calculated this way will be less than the similarity if there was a path between them in the context region. Note that this evolution process is only used to compute the similarity between the concepts and does not reflect as the change in the user context.
5. A PHOTO RETRIEVAL APPLICATION
We applied our context model in a photo retrieval system to improve the retrieval efficiency. Retrieval of photographs in a large personal collection is an important problem due to the easy availability of image capturing devices. We now discuss the architecture and different components of the photo retrieval application.
5.1 Architecture of the application
The system block diagram (ref. Figure 4) shows three main blocks namely the user interaction, the context-aware search engine and the context model as discussed earlier. The application is multi-threaded for efficiency – the three main blocks operate as three different threads.
The context model runs as a background thread while the User Interaction is the main thread. The search thread is launched when the user makes a query and ends when the results are obtained. The context model consists of the three different knowledge spaces namely the User Knowledge, the Environment Knowledge and the Media Knowledge. The User Knowledge is dynamic while the Media Knowledge and the Environment Knowledge are static. The user interactions provide information about the user and this information is used by the context model to update the User knowledge and the User context. The context evolution engine performs this task. The search engine uses the current user context and the media knowledge to provide the retrieval results. We now discuss the user interaction and the search engine.
5.2 User Interaction in the System
The user interacts with the system through an applet shown in Figure 5. In order to provide a text query, the query is typed in the text box in the upper left hand side of the screen. The displayed images can be selected by clicking the check box with the images. Finally the query is submitted by clicking the submit query button. The top nine results are displayed in the applet with the top most result also shown alongside. It has been studied that the human short term memory can efficiently process seven plus minus two units of information at a time [22], so we chose nine images to display as results.
The user time spent on the displayed pictures and the selection of pictures are the messages sent by the user to the system. These messages provide information about the
current user activity and are used by the evolution engine to update the context.
5.3 Context-Aware Search in the Application
Given the query as a set of selected images, the context-aware search is performed in the media knowledge space to find the most relevant photographs. The search process first obtains the current context from the context model and modifies it using the user information obtained from the query. The modified context is then used to obtain the candidate concepts in the media knowledge space. The images close to the candidate concepts in the media knowledge space form the retrieval results. The complete search is as follows:
Find the selected concepts from the selected
images (the color histogram of the images and the annotations).
Expand the user knowledge if the selected
concepts are not present in the user knowledge. Evolve the user context with activation at the
selected concepts. This evolution process is done by the search thread and does not reflect in the new user context model. The concepts (both images and text) in the user context that are also present in the media knowledge and have weight greater than α (optimized experimentally) of the highest weight form the set CC of candidate concepts.
The score of an image in the media knowledge is
now given as
()max{*(,)}c cc k k
S l w S l CC =, <11>
where CC k is the k th
candidate concept and the S (l , CC k ) is the similarity between the image and the candidate concept k in the media knowledge. The similarity is computed as in the user knowledge space as discussed in section 4.6. Nine images with highest scores are the result images. The displayed images form the new information viewed by the user. These images and their annotations are set as the current user information for the context evolution.
6. EXPERIMENTS AND EVALUATION
We now discuss the user experiments with the Photo retrieval application. We conducted a pilot user study with six graduate students. The experiments were conducted to demonstrate the effectiveness of the context-aware retrieval and hence the focus was not on using the optimal feature set.
The media repository was a set of 4000 images from the users’ personal and shared photograph collection. Roughly 15% of these images were text annotated. Personal photographs makes the experiments more complex and better approximate real-world usage scenarios than the familiar Corel data set. Also importantly, the semantic associations are different for different users and we expect
them to be captured by the User knowledge estimates.
Figure 4: Block diagram of the photo retrieval
application
The media retrieval application was implemented in Java and executed on Pentium 4 machine. The program structure is as discussed in section 4.
6.1Three Retrieval strategies
We implemented three retrieval strategies namely random retrieval, relevance feedback based retrieval and context-based retrieval.
Random: In the random retrieval scenario, the images in the display set were selected at random.
This strategy was selected as a baseline against
which the other retrieval strategies were
compared. The retrieval precision for this strategy
should be close to the ratio of images in the
database that are relevant to the query concept and
is the minimum precision expected of any good
retrieval strategy.
Relevance Feedback: The relevance feedback based retrieval strategy was implemented as
defined in the algorithm discussed in [27]. The
color histograms of the images were the selected
features for the experiments and the weights of the
dimensions and similarities between the images
were computed as discussed in [27]. However, for
the consistency of the experiment and comparison
with other approaches, we allowed only two levels
of relevance score for each retrieved item. Thus a
retrieved image is given a relevance score 1 if the
user selects it as a relevant image. All other
images in the database are given a relevance score
0.
Our Algorithm: The context-based retrieval strategy used the context model and retrieved the
images as discussed in section 4.
Thus, three experimental setups were made, each implementing one of the three retrieval strategies. The user interface and interaction instruction for the three setups were same for transparency to the user. 6.2Description of Experiments
We now discuss the experimental set-up and the instructions given to the user. We required that each user provide a set of at least ten concepts. This set of concepts was the seed with which the user knowledge and context were initialized as discussed in section 4.3. Then each user was asked to follow the instructions given below:
Select a query concept from among a set of choice concepts. These choice concepts were selected
from the list of concepts in the media knowledge
that had at least 100 relevant images. The first set
of 9 images was selected randomly. We set the
number of images to be nine due to the familiar
property of short term memory [22].
Once the images were presented, the users were asked to select among them, the images that were
relevant, by selecting the associated checkboxes.
The selected images were used as the query and the new images were presented. The images once
shown are not repeated in the same set-up. This
process is repeated four times. Thus the users see
a total of 45 images.
Each user repeated the above experiment three times. Each time one of the three different search strategies namely, random retrieval, relevance feedback based retrieval and context-aware retrieval was used. The order of the experiments for each users was randomly selected and the order was not disclosed to the users.
6.3Observations and Discussions
We now analyze the experimental results as both cumulative precision of the overall retrieved set and the change in the relevance score with increasing interaction and the personal priorities of different users.
6.3.1Cumulative Precision
Table 1: Number of retrieved images for different queries and the % of relevant images in the database.
Number of retrieved images in setup Query
% in
database Random Relevance
Feedback
Context Home 10 6 12
14
Birthday 10 4 14 12
Graduation 3 1 9 14
Beach 5 2 15 16
Park 20 8 20 23
Office 5 1 8 10
We present the cumulative precision results as the number
of relevant images that were retrieved in the complete
experiment of five iterations and the mean relevance score
of the retrieved images in the five iterations. The number of
relevant images obtained in five iterations for three
different search strategies are shown in Table 1.
Figure 5: The user interface to the search application
We observe that the context-based retrieval gives the largest number of images relevant to the query. The number of relevant images retrieved using the random retrieval strategy depends solely on the percentage of relevant images in the database. It can be seen that the random retrieval strategy gives more number of relevant images for the query ‘park’ (20% images) as compared to the query ‘beach’ (5% images). The difference between the context-based retrieval and the relevance feedback method is not seen clearly as we do not consider the rank of the images.
A more revealing analysis of the results is possible by considering the rank of the retrieved images. The normalized relevance score for the retrieved set of nine images is computed as follows:
1
2
(1)(1)N
j j S N r N N ==+?+∑, <12>
where N is the number of images and r i represent the rank of the i th image given as
is relevant to query 0otherwise i i i r ?=??
<13>
Equation <12> helps us distinguish between two returned
sets that have identical precision, as it takes the rank into account. The mean cumulative relevance score of the results for all the five iterations is shown in Table 2. The mean relevance score of the retrieved set for the context-based retrieval is significantly larger than the relevance feedback approach for all the queries. This demonstrates that context-based retrieval is more effective than the relevance feedback retrieval.
Table 2: Mean relevance score of the retrieved images for
different queries
Mean relevance score of retrieved image set Query
% in database
Random
Relevance Feedback Context Home 10 0.07 0.29 0.32 Birthday 10 0.07 0.24 0.38 Graduation 3
0.02
0.21
0.34
Beach 5
0.03
0.28
0.42
Park 20
0.16
0.44
0.51 Office 5
0.02
0.13 0.29
We now discuss the change in the relevance of the results with increasing user interaction.
6.3.2 Change in relevance with interaction
An important aspect of the context-based retrieval approach is that with increasing interaction more relevant images are retrieved. The relevance scores in different iterations for the three different search strategies are shown in Figure 6. The relevance score of the retrieved set is seen increasing with the increasing interaction in the context-based retrieval strategy. The relevance feedback based
approach also shows and increasing trend but is not very consistent.
There are two reasons for the improved dynamic performance of the context-based approach against any other approach. Firstly, with increasing interaction, the estimate of the user context becomes more accurate and the
retrieved set is therefore more relevant. Secondly, the context-based approach uses multi-modal relations (commonsensical relations using the annotations and feature based metrics) between concepts to estimate the similarities. This helps retrieving images that are more semantically similar to the query. We see an anomaly in bottom figure in Figure 6, where in the fourth iteration; the relevance score of the relevance feedback approach is higher than the context-based retrieval approach. This is an outlier and is ascribed to the human error in interaction.
7. LIMITATIONS
The use of the context model provided significant improvement in the photo retrieval over relevance feedback
Figure 6: Plot of relevance score against user
interaction for queries ‘home’ (top) and ‘park’ (bottom)
queries.
base retrieval however, the model has the following limitations:
The quality of the acquired user knowledge. For experimental purposes, we have used the concept
net [20] to extract textual concepts and the
commonsensical relationships between them while
image concepts are derived from the media
database itself. The textual concepts and their
relationships in concept-net are not very dense and
that limits the knowledge.
The model needs continuous interaction with the user to have an accurate estimate of the user. In
the absence of interaction, though the model
maintains the knowledge about the user, until the
past interactions, the estimate of the current
context is very conservative.
From the perspective of knowledge representation, while we provided a generic representation
framework for knowledge in all modalities, the
knowledge used in the application was limited to
only text and image planes. In future work, we are
planning to incorporate representations of human
activities. Our current choice is guided by the
needs of the image retrieval application.
Knowledge is representable at multiple time-scales. We did not cover the multi-scale temporal
nature of knowledge in our current framework,
and plan to incorporate this into future work.
Probabilistic: The presented framework does not incorporate a probabilistic knowledge
representation framework. We are currently
developing new statistical models to represent
multi-sensory knowledge and user action context
using partially observable decision markov
processes [18,32].
We plan to address these limitation in future work.
8.CONCLUSIONS
In this paper, we presented a novel framework for modeling user context. We have developed a theoretical understanding of multi-sensory knowledge and user context and their inter-relationships. We presented our definition of context as “the finite and dynamic set of multi-sensory and inter-related conditions that influences the exchange of messages between two entities in communication.” We stated properties of context and its relationship to knowledge and the application. We then presented a dynamic, emergent and multi-sensory knowledge representation framework in which context can
be estimated and represented as a subset that is in focus. We modeled the dynamics of the context and knowledge as linear dynamical system. We provided proofs for the convergence, stability and controllability of the user context model under all circumstances. We tested the validity of our proposed user context model by applying it to a photo retrieval application. Our experiments demonstrate that the model of user context helps retrieve more relevant images for the user as compared to relevance feedback. The experiments also demonstrate the improvement in the relevance of the retrieved images with increasing user interaction.
We have identified that for a robust context model, we need a comprehensive framework for knowledge acquisition and representation. Hence, we plan to focus our future work mainly on improving the process of user knowledge acquisition, and multi-scale and multi-modal knowledge representation. We are also working on more exhaustive experiments with more users and a larger dataset and improved features to provide statistically more significant results.
9.REFERENCES
[1] ConceptNet
https://www.doczj.com/doc/9c7079709.html,/~hugo/conceptnet.
[2] OpenCyc https://www.doczj.com/doc/9c7079709.html,.
[3] P.A PPAN and H.S UNDARAM (2004). Networked
multimedia event exploration, Proc. ACM Multimedia
2004, also AME TR-2004-10, pp. 40-47, Oct. 2004.,
New York, New York.
[4] R.A TKINSON and R.S HIFFRIN (1968). Human
memory: A proposed system and its control processes.
The psychology of learning and motivation: Advances
in research and theory(eds). New York, Academic
Press.
[5] A.B.B ENITEZ, J.R.S MITH and S.-F.C HANG (2000).
MediaNet: A Multimedia Information Network for
Knowledge Representation, Proceedings of the 2000
SPIE Conference on Internet Multimedia
Management Systems (IS&T/SPIE-2000), Nov 6-8,
2000., Boston MA.
[6] C.-T.C HEN (1999). Linear Systems Theory and
Design. Oxford University Press.
[7] W.C HRISTENSEN (2004). Self-directedness,
integration and higher cognition. Language Sciences
26(6): 661-692.
[8] I.J.C OX, M.L.M ILLER, T.P.M INKA, T.V.
P APATHOMAS and P.N.Y IANILOS (2000). The
Bayesian Image Retrieval System, PicHunter: Theory,
Implementation and Psychophysical Experiments.
IEEE Trans. Image Processing---Special Issue on
Digital Libraries.
[9] A.K.D EY (2000). Providing Architectural Support
for Building Context-Aware Applications. College of
Computing. Atlanta, Georgia Institute of Technology, PhD.
[10] A.K.D EY (2001). Understanding and Using Context.
Personal and Ubiquitous Computing Journal 5(1): 4-7.
[11] A.K.D EY and G.D.A BOWD (1999). Towards a
Better Understanding of Context and Context-
Awareness, Proceedings of the 3rd International
Symposium on Wearable Computers, pp. 21-28,
October 20-21, 1999, San Francisco, CA.
[12] A.K.D EY, M.F UTAKAWA, D.S ALBER and G.D.
A BOWD (1999). The Conference Assistant: Combining
Context-Awareness with Wearable Computing,
Proceedings of the 3rd International Symposium on
Wearable Computers, pp. 21-28, October 20-21,
1999., San Francisco, CA.
[13] P.D OURISH (2004). What we talk about when we talk
about context. Personal and Ubiquitous Computing
8(1): 19-30.
[14] S.G REENBERG (2001). Context as a Dynamic
Construct. Human-Computer Interaction 16: 257-268.
[15] E.H YVONEN, S.S AARELA and K.V ILJANEN (2003).
Ontogator: Combining View- and Ontology-Based
Search with Semantic Browsing, Proc. of XML,
October 30-31, 2003, Kuopio, Finland.
[16] R.J AKOBSON (1960). Closing statement: linguistics
and poetics. Style in Language. T.S EBEOK. (eds).
Cambridge MA, MIT: 350-377.
[17] H.L IEBERMAN and T.S ELKER (2000). Out of Context
: Computer Systems that Adapt To, and Learn From,
Context. IBM Systems Journal 39(3,4): 617-631. [18] M.L.L ITTMAN, A.R.C ASSANDRA and L.P.
K AELBLING (1995). Learning Policies for Partially
Observable Environments: Scaling Up, Proceedings
of the Twelfth International Conference on Machine
Learning, July 1995, Lake Tahoe, CA.
[19] H.L IU and H.L IEBERMAN (2002). Robust Photo
Retrieval Using World Semantics, LREC, Las Palmas, Canary Islands.
[20] H.L IU and P.S INGH (2004). ConceptNet: a practical
commonsense reasoning toolkit. BT Technology
Journal 22(4): pp. 211-226.
[21] S.M AKARIOS and R.G UHA (2005). A First Order
Theory of Contexts, Context 2005, Paris.
[22] G.A.M ILLER (1956). The magical number seven, plus
or minus two: Some limits on our capacity for
processing information. Psychological Review 63: 81-
97.
[23] G.A.M ILLER, R.B ECKWITH and C.F ELLBAUM (1993).
Introduction to WordNet : An on-Line Lexical
Database. International Journal of Lexicography 3(4): 235-244. [24] K.M URPHY and W.F REEMAN (2004). Contextual
Models for Object Detection using Boosted Random
Fields, NIPS'04,
[25] D.O'SULLIVAN, E.M C L OUGHLIN, M.B ERTOLOTTO
and D.W ILSON (2005). Context-Oriented Image
Retrieval, Context 2005, Paris.
[26] J.-C.P OMEROL and P.B REZILLON (2001). About some
relationships between Knowledge and Context,
Context-01, Dundee, Scotland.
[27] Y.R UI and T.H UANG (1999). A Novel Relevance
Feedback Technique in Image Retrieval., Proc. ACM
Multimedia 1999, Nov. 1999, Orlando, FL.
[28] B.N.S CHILIT and M.M.T HEIMER (1994).
Disseminating active map information to mobile
hosts. IEEE Network 8(5): 22-32.
[29] B.S HEVADE and H.S UNDARAM (2004). Incentive
Based Image Annotation. Arts Media and Engineering Program, Arizona State University, AME-TR-2004-
02, Jan. 2004.
[30] A.W.M.S MEULDERS, M.W ORRING, S.S ANTINI, A.
G UPTA and R.J AIN (2000). Content-based image
retrieval at the end of the early years. IEEE
Transactions on Pattern Analysis and Machine
Intelligence 22(12): 1349-1380.
[31] L.A.S UCHMAN (1987). Plans and situated actions :
the problem of human-machine communication.
Cambridge University Press Cambridge
Cambridgeshire ; New York.
[32] G.T HEOCHAROUS, K.M URPHY and L.P.K AELBLING
(2003). Representing hierarchical POMDPs as DBNs for multi-scale robot localization, Workshop on
Reasoning about Uncertainty in Robotics,
International Joint Conference on Artificial
Intelligence, Acapulco, Mexico.
[33] S.V EERAMACHANENI, P.S ARKAR and G.N AGY
(2005). Modeling Context as Statistical Dependence,
Context 2005, Paris.
[34] T.W INOGRAD (2001). Architectures for Context.
https://www.doczj.com/doc/9c7079709.html,/~winograd/papers/context/cont
ext.pdf.
[35] G.W U, E.Y.C HANG and N.P ANDA (2005).
Formulating Context-dependent similarity functions,
ACM Multimedia, 725-734, Singapore.
两招让电脑自动关机的方法 经常使用迅雷的朋友，一定知道“下载完成自动关机”的功能吧，但有些软件，或者说在电脑上自带的软件没有此功能，大家就要等着下载完成然后关机，多费时，其实在电脑上有自动关机的命令。 一、设置电脑自动关机命令 1.开始—运行—输入“at 12:00 Shutdown -s”,这样电脑在12点钟就会出现“系统关机”对话框，默认有30秒的倒计时并提示保存; 运行 2.开始—运行—输入“Shutdown –s –t 3600”,这里“3600”是表示60分钟，代表的是60分钟之后自动关机。 注：如果想取消的话，在运行中输入“shutdown -a”即可。 二、电脑自动关机软件 1.直接在网上搜索自动关机软件，会出来多款，自己选择中意的一款下载即可。 2.类似的关机软件是免费上网共享关机软件，有些关机软件不但具有定时自动关机的功能，而且还可以进行定时提醒信息、打开文件、打开网页、重启电脑、注销用户、锁定计算机、计算机休眠或者待机等等，并且可以多任务进行，可以满足不同用户的要求，禁止行为功能可以让电脑防止他人的恶意破坏。 3.要是经常设置自动关机的用户，小编就建议使用第二种自动关机软件，比较方便快捷。第一种方式设置电脑，存在一个问题就是有的系统(如win7)在行使命令时，需要在管理员的权限下运行，这就需要用户另外设置菜单，具体方法如下：点击“开始”，在“搜索程序和文件” 框中输入“cmd.exe”，在输出结果列表中找到cmd.exe，鼠标右键单击，在菜单中选择“以管理员身份运行”。
以上就是小编为大家推荐的让电脑自动关机的两种方法，电脑自动关机命令的设置，在“运行”中输入相关命令既可以设置完成。另一种方法就是借助自动关机软件，在软件中修改设置也是可以办得到的。
怎样设置电脑到时间会自动关机和开机 Windows XP的关机是由Shutdown.exe程序来控制的，位于Windows\System32文件夹中。如果想让Windows 2000也实现同样的效果，可以把Shutdown.exe复制到系统目录下。 比如你的电脑要在22:00关机，可以选择“开始→运行”，输入“at 22:00 Shutdown -s”，这样，到了22点电脑就会出现“系统关机”对话框，默认有30秒钟的倒计时并提示你保存工作。如果你想以倒计时的方式关机，可以输入“Shutdown.exe -s -t 3600”，这里表示60分钟后自动关机，“3600”代表60分钟。 设置好自动关机后，如果想取消的话，可以在运行中输入“shutdown -a”。另外输入“shutdown -i”，则可以打开设置自动关机对话框，对自动关机进行设置。 Shutdown.exe的参数，每个都具有特定的用途，执行每一个都会产生不同的效果，比如“-s”就表示关闭本地计算机，“-a”表示取消关机操作，下面列出了更多参数，大家可以在Shutdown.exe中按需使用 电脑自动开机并播放音乐提醒你起床,并且晚上自动关机 使用电脑的自动开机,哇,好爽,并且音乐可以随便你定哦，一天都有好的心情. 1.自动定时开机此功能需要在BIOS中设定,并且你的主板的高级电源管理设置必须支持这个功能,不过现在大多数的主板全支持,放心使用吧. 步骤一: 启动系统按DEL,进入BIOS,然后选中"POWER MANAGEMENT FEATURES" (电源管理设置),回车进入.在电源窗口中找到"RESTORE ON AC/POWER LOSS"项,按回车,然后通过方向键设置其值为"POWER ON". 步骤二: 选中"RESUME ON RTC ALARM"并按回车,接着通过方向键设置其值为"ENABLE" 步骤三: 选中"RTC ALARM DATE"按回车,然后将其值设定为"EVERYDAY",表示每天都进行一样的操作,按照同样的方法,分别将"RTC ALARM HOURS""RTC ALARM MINUTE""RTM ALARM SECOND"设置为"07""15""00",这表示开机时间为早上的"7:15:00",全部设置好后,按F10,保存并重新启动. (由于不同的主板的BIOS不一样,有些主板的设置可能与上面的有一点点不同，但只要记住在电源管理中找到"WAKE UP"字样的就行了，因为这是设置定时开机的选项) 2自动登陆系统电脑开机了，但是我们有些人设置了开机密码,不能自动登陆,怎么办? 我们可以通过修改注册表来实现自动登陆. 步骤一: 在运行中输入"REGEDIT",打开HKEY_LOCAL_MACHINE\SOFTWARE\MICROSOFT\WINDOWSNT\CURRENTVERSIO N\WINLOGON. 步骤二: 在右方的窗口中找到"DEFAULTUSERNAME"子键,将数值数据改
[摘要] 计算机网络安全建设是涉及我国经济发展、社会发展和国家安全的重大问题。本文结合网络安全建设的全面信息，在对网络系统详细的需求分析基础上，依照计算机网络安全设计目标和计算机网络安全系统的总体规划，设计了一个完整的、立体的、多层次的网络安全防御体系。 [关键词] 网络安全方案设计实现 一、计算机网络安全方案设计与实现概述 影响网络安全的因素很多，保护网络安全的技术、手段也很多。一般来说，保护网络安全的主要技术有防火墙技术、入侵检测技术、安全评估技术、防病毒技术、加密技术、身份认证技术，等等。为了保护网络系统的安全，必须结合网络的具体需求，将多种安全措施进行整合，建立一个完整的、立体的、多层次的网络安全防御体系，这样一个全面的网络安全解决方案，可以防止安全风险的各个方面的问题。 二、计算机网络安全方案设计并实现 1.桌面安全系统 用户的重要信息都是以文件的形式存储在磁盘上，使用户可以方便地存取、修改、分发。这样可以提高办公的效率，但同时也造成用户的信息易受到攻击，造成泄密。特别是对于移动办公的情况更是如此。因此，需要对移动用户的文件及文件夹进行本地安全管理，防止文件泄密等安全隐患。 本设计方案采用清华紫光公司出品的紫光S锁产品，“紫光S锁”是清华紫光“桌面计算机信息安全保护系统”的商品名称。紫光S锁的内部集成了包括中央处理器（CPU）、加密运算协处理器（CAU）、只读存储器（ROM），随机存储器（RAM）、电可擦除可编程只读存储器（E2PROM）等，以及固化在ROM内部的芯片操作系统COS（Chip Operating Sys tem）、硬件ID号、各种密钥和加密算法等。紫光S锁采用了通过中国人民银行认证的Sm artCOS，其安全模块可防止非法数据的侵入和数据的篡改，防止非法软件对S锁进行操作。 2.病毒防护系统 基于单位目前网络的现状，在网络中添加一台服务器，用于安装IMSS。
电脑自动关机 电脑自动关机是计算机的一个命令。常见的有windows的自动关机，还有嵌入式系统手机系统的自动关机。 使用“shutdown”可以让电脑自动关机，不过如果打开很多文档就会提示你保存文件不，你不去手动确定，shutdown这个命令是没用的。比如你的电脑要在22:00关机，可以选择“开始→运行”，输入“at 22:00 Shutdown -s”，这样，到了22点电脑就会出现“系统关机”对话框，默认有30秒钟的倒计时并提示你保存工作。如果你想以倒计时的方式关机，可以输入“Shutdown.exe -s -t 3600”，这里表示60分钟后自动关机，“3600”代表60分钟。 设置好自动关机后，如果想取消的话，可以在运行中输入“shutdown -a”。另外输入“shutdown -i”，则可以打开设置自动关机对话框，对自动关机进行设置。 Shutdown.exe的参数，每个都具有特定的用途，执行每一个都会产生不同的效果，比如“-s”就表示关闭本地计算机，“-a”表示取消关机操作，下面列出了更多参数，大家可以在Shutdown.exe中按需使用。 -f：强行关闭应用程序 -m \\计算机名：控制远程计算机 -i：显示图形用户界面，但必须是Shutdown的第一个选项 -l：注销当前用户 -r：关机并重启 -t 时间：设置关机倒计时 -c "消息内容"：输入关机对话框中的消息内容（不能超127个字符） 非正常自动关机的故障原因 病毒问题 有黑客通过系统漏洞对你的电脑进行入侵，然后就可以远程控制你的电脑所有行为，包括电脑关机,可以使用杀毒软件对电脑杀毒和打系统补丁。 系统文件被破坏 如果操作系统重要的运行文件丢失或者损坏，电脑自启动时候会因此无法完成初始化导致自动关机。 硬件问题引起的 例如电源电压不稳定，插排或者电源插头接触不良，散热不良，cpu，内存，光驱损坏。
校园网络与信息安全管理办法 桑梓镇辛撞中心小学 2018.9
校园网络与信息安全管理办法 学校校园网是为教学及学校管理而建立的计算机信息网络，目的在于利用先进实用的计算机技术和网络通信技术，实现校园内计算机互联、资源共享，并为师生提供丰富的网上资源。为了保护校园网络系统的安全、促进学校计算机网络的应用和发展，保证校园网络的正常运行和网络用户的使用权益，更好的为教育教学服务，特制定如下管理办法。 第一章总则 1、本管理制度所称的校园网络系统，是指由校园网络设备、配套的网络线缆设施、网络服务器、工作站、学校办公及教师教学用计算机等所构成的，为校园网络应用而服务的硬件、软件的集成系统。 2、校园网络的安全管理，应当保障计算机网络设备和配套设施的安全，保障信息的安全和运行环境的安全，保障网络系统的正常运行，保障信息系统的安全运行。 3、按照“合法合规，遵从标准”的原则开展网络与信息安全管理工作，网络及信息安全管理领导小组负责相应的网络安全和信息安全工作，定期对相应的网络用户进行有关信息安全和网络安全教育并对上网信息进行审查和监控。 4、所有上网用户必须遵守国家有关法律、法规，严格执行安全保密制度，并对所提供的信息负责。任何单位和个人不得利用联网计算机从事危害校园网及本地局域网服务器、工作站的活动。 5、进入校园网的全体学生、教职员工必须接受并配合国家有关部门及学校依法进行的监督检查，必须接受学校校园网络及信息安全管理领导小组进行的网络系统及信息系统的安全检查。
6、使用校园网的全体师生有义务向校园网络及信息安全管理领导小组和有关部门报告违法行为和有害信息。 第二章网络安全管理细则 1、网络管理中心机房及计算机网络教室要装置调温、调湿、稳压、接地、防雷、防火、防盗等设备，管理人员应每天检查上述设备是否正常，保证网络设备的安全运行，要建立完整、规范的校园网设备运行情况档案及网络设备账目，认真做好各项资料(软件)的记录、分类和妥善保存工作。 2、校园网由学校信息中心统一管理及维护。连入校园网的各部门、处室、教室和个人使用者必须严格使用由信息中心分配的IP地址，网络管理员对入网计算机和使用者进行及时、准确登记备案，由信息中心负责对其进行监督和检查。任何人不得更改IP及网络设置，不得盗用IP地址及用户帐号。 3、与校园网相连的计算机用户建设应当符合国家的有关标准和规定，校园内从事施工、建设，不得危害计算机网络系统的安全。 4、网络管理员负责全校网络及信息的安全工作，建立网络事故报告并定期汇报，及时解决突发事件和问题。校园网各服务器发生案件、以及遭到黑客攻击后，信息中心必须及时备案并向公安机关报告。 5、网络教室及相关设施未经校领导批准不准对社会开放。 6、按照信息安全等级保护工作规定，完成定级、备案等工作，留存安全审核日志。校园网中对外发布信息的Web服务器中的内容必须经领导审核，由负责人签署意见后再由专人（信息员）发布。新闻公布、公文发布权限要经过校领导的批准。门户网站不得链接企业网站，不得发布商业广告，不得在网页中设置或植入商品、商业服务的二维码。
Wind7自动关机命令 相信大家对电脑自动关机的概念都不陌生吧，现在小编与大家携手探讨电脑高手都知道的一些Win7自动关机命令。首先，我们需要做如下步骤：1.“开始”→“运行”→“cmd”→输入你想要设置的具体自动关机时间，接下来你就可以做照着以下步骤做就可以了。 设置1定时自动关机 假如你需要电脑在20:00自动关机，那么只需要在“开始”菜单中选择“运行”，然后输入at 20:00 Shutdown -s，按“Enter”键即可，但是你必须先启动Task Scheduler服务。这样当电脑系统时间到了20:00时，电脑就会自动出现“系统关机”的对话框，30秒以后就会自动关机，这30秒是为了让用户在关机前保存现在运行的工作，以免造成不必要的损失，提高了安全性。（-s参数是指关闭此计算机命令） 设置2倒计时自动关机 假如你需要电脑在10分钟以后自动关机，那么只需要在“开始”菜单中选择“运行”，然后输入Shutdown -s -t 600，按“Enter”键即可。这样电脑即开始倒计时，10分钟以后电脑将会自动关机。（-t参数是设置关闭此计算机的倒计时命令，时间是以秒为单位的） 设置3取消自动关机 假如你已经设置了自动关机，但又改变了主意，不必自动关机了，那么只需要在“开始”菜单中选择“运行”，然后输入Shutdown -a，点击“确定”即可。这样即关闭了已经设置的自动关机。（-a参数是取消自动关机命令） 设置4定时重启 假如你需要电脑在20:00重新启动，那么只需要在“开始”菜单中选择“运行”，然后输入at 20:00 Shutdown -r，点击“确定”即可。这样当电脑系统时间到了20:00时，电脑就会自动重新启动了。（-r参数是关闭并重新启动此计算机命令） 设置5倒计时重启 假如你需要电脑在10分钟以后重新启动，那么只需要在“开始”菜单中选择“运行”，然后输入。 Shutdown -r -t 600，点击“确定”即可。这样电脑即开始倒计时，10分钟以后电脑将会重新启动 另外，小编还整理了以下命令供大家选择使用： -f：强行关闭应用程序 -m \\computer：控制远程计算机 -i：显示图形用户界面，但必须是Shutdown的第一个选项 -l：注销当前用户 -c "消息内容"：输入关机对话框中的消息内容（不能超127个字符） 此外，相信大家也很想了解远程开机方法，小编向大家推荐一款远程开机软件-网络人电脑控制器，是南宁网络人电脑有限公司出品的一款远程开机产品，有固话和手机两个版本。安装时将控制器插在电脑的PCI 插槽上，不需要其他软件配合或在电脑上安装驱动。固话版需要从电话机上分出一根电话线插到控制器上，手机版需要在控制上放置一张手机卡。控制器安装好后，只要拨打该电话或手机号，即可实现远程开机。
校园网络安全设计方案 10网工2班组员：张婵、张茜、张越、张喻博、赵子龙、祝美意、杨越峦、张力 随着因特网的迅速发展，校园网的建设日益普遍。而在高校中，如何能够保证校园网络的安全运行，同时又能提供丰富的网络资源，达到办公、教学及学生上网的多种需求已成为了一个难题。校园网络的安全不仅有来自外部的攻击，还有内部的攻击。所以，在校园网建设中使用安全技术是刻不容缓的。现从防火墙、VPN、防病毒、入侵检测和防御系统、上网行为管理和用户审计系统、数据备份系统、主页防篡改、网络安全管理制度几个方面，设计我校的网络安全方案。 防火墙:防火墙是一种将内部网和公众网分开的方法。它能限制被保护的网络与与其他网络之间进行的信息存取、传递操作。 防火墙的概念：通常防火墙是指部署在不同网络或网络安全域之间的一系列部件组合，是一种有效的网络安全策略。防火墙提供信息安全服务，设置在被保护内部网络的安全与不安全的外部网络之间，其作用是阻断来自外部的、针对内部网络的入侵和威胁，保护内部网络的安全。它是不同网络或网络安全域之间信息的唯一出入口，根据安全策略控制出入网络的信息流，并且本身具有较强的抗攻击能力。 防火墙的分类：按软件与硬件的形式，防火墙分为软件防火墙、硬件防火墙和芯片防火墙；按防火墙的技术，总体分为包过滤型和应用代理型两大类；按防火墙的结构分为单一主机防火墙、路由器集成式防火墙、分布式防火墙；按防火墙的部署位置分为边界防火墙、个人防火墙、混合防火墙。 防火墙的安全策略：（1）所有从内到外和从外到内的数据包都必须经过防火墙（2）只有被安全策略允许的数据包才能通过防火墙（3）防火墙本身要有预防入侵的功能（4）默认禁止所有服务，除非是必须的服务才被允许 防火墙的设计：（1）保障校园内部网主机的安全，屏蔽内部网络，禁止外部网用户连接到内部网（2）只向外部用户提供HTTP、SMTP和POP等有限的服务（3）向内部记账用户提供所有Internet服务，但一律通过代理服务器(4)禁止访问黄色、反动网站(5)要求具备防IP 地址欺骗和IP地址盗用功能(6)要求具备记账和审计功能，能有效记录校园网的一切活动。 校园网络在设置时应从下面几个方面入手：（1）入侵检测：具有黑客普通攻击的实时检测技术。实时防护来自IP Source Routing、IP Spoofing、SYN flood、IC-MP flood、UDP flood、Ping ofDeath、拒绝服务和许多其它攻击。并且在检测到有攻击行为时能通过电子邮件或其它方式通知系统管理员。（2）工作模式选择：目前市面上的防火墙都会具备三种不同的工作模式，路由模式、NAT模式和透明模式。我们选择的是透明模式，防火墙过滤通过防火墙的封包，而不会修改数据包包头中的任何源或目的地的信息。所有接口运行起来都像是同一网络中的一部分。此时防火墙的作用更像是Layer2（第二层）交换机或桥接器。在透明模式下，接口的IP地址被设置为0.0.0.0, 防火墙对于用户来说是可视或透明的。（3）策略设置：防火墙可以提供具有单个进入和退出点的网络边界。由于所有信息流都必须通过此点，因此可以筛选并引导所有通过执行策略组列表产生的信息流。策略能允许、拒绝、加密、认证、排定优先次序、调度以及监控尝试从一个安全段流到另一个安全段的信息流。可以决定哪些用户和信息能进入和离开以及它们进入和离开的时间和地点。（4）管理界面：管理一个防火墙的方法一般来说有两种：图形化界面和命令行界面，我们选择为通过web方式和java等程序编写的图形化界面进行远程管理。（5）内容过滤：面对当前互联网上的各种有害信息，我们的防火墙还增加了URL阻断、关键词检查、Java Ap-ple、ActiveX和恶意脚本过滤等。（6）防火墙的性能考虑：防火墙的性能对于一个防火墙来说是至关重要的，它决定了每秒钟可能通过防火墙的最大数据流量，以bps为单位，从几十兆到几百兆不等。千兆防火墙还
台式电脑设置自动关机的方法 很多网友都在问说台式电脑如何设置自动关机，那么下面就由小编来给你们说说台式电脑设置自动关机的方法吧，希望可以帮到你们哦! 台式电脑设置自动关机的方法一： 按“win+R”(注win键即指windows徽标键，该键的最明显的标识就是在键面上有个Windows的徽标)，此时会弹出“运行”对话框，如下图所示。 在对话框的编辑栏里输入“Shutdown -s -t 600”(注：引号不输入，“-s”“-t”及“600”前面均有一个空格，其中的数字代表的是时间，单位为秒，如上600即代表10分钟)，如下图所示。
然后点击“确认”，完成设置。此时，系统会弹出系统提示对话框，提示系统将在10分钟内关闭，如下图所示。 台式电脑设置自动关机的方法二： 设置一定时自动关机 假如你需要电脑在20：20自动关机，那么只需要在“开始”菜单中选择“运行”，然后输入at 20:20 Shutdown -s，点击“确定”即可。这样当电脑系统时间到了20：20时 电脑就会自动出现“系统关机”的对话框，30秒以后就会
自动关机，这30秒是为了让用户在关机前保存现在运行的工作，以免造成不必要的损失。 注意：-s参数是指本地计算机。 设置二倒计时自动关机 假如你需要电脑在10分钟以后自动关机，那么只需要在“开始”菜单中选择“运行”，然后输入Shutdown -s -t 600，点击“确定”即可。这样电脑即开始倒计时，10分钟以后电脑将会自动关机。 注意：-t参数是倒计时自动关机的命令，600参数是10分钟即600秒。 设置三取消自动关机 假如你已经设置了自动关机，但又改变了主意，不必自动关机了，那么只需要在“开始”菜单中选择“运行”，然后输入Shutdown -a，点击“确定”即可。这样即关闭了已经设置的自动关机。 注意：-a参数是取消自动关机的命令。
校园网网络安全设计方案 网络技术的高速发展,各种各样的安全问题也相继出现,校园网被“黑”或被病毒破坏的事件屡有发生,造成了极坏的社会影响和巨大的经济损失。维护校园网网络安全需要从网络的搭建及网络 安全设计方面着手。 一、基本网络的搭建。 由于校园网网络特性(数据流量大,稳定性强,经济性和扩充性)和各个部门的要求(制作部门和办公部门间的访问控制),我们采用下列方案: 1. 网络拓扑结构选择:网络采用星型拓扑结构(如图1)。它是目前使用最多,最为普遍的局域网拓扑结构。节点具有高度的独立性,并且适合在中央位置放置网络诊断设备。 2.组网技术选择:目前,常用的主干网的组网技术有快速以太网(100Mbps)、FDDI、千兆以太网(1000Mbps)和ATM(155Mbps/622Mbps)。快速以太网是一种非常成熟的组网技术,它的造价很低,性能价格比很高;FDDI也是一种成熟的组网技术,但技术复杂、造价高,难以升级;ATM技术成熟,是多媒体应用系统的理想网络平台,但它的网络带宽的实际利用率很低;目前千兆以太网已成为一种成熟的组网技术,造价低于ATM网,它的有效带宽比622Mbps的ATM 还高。因此,个人推荐采用千兆以太网为骨干,快速以太网交换到桌面组建计算机播控网络。 二、网络安全设计。 1.物理安全设计为保证校园网信息网络系统的物理安全,除在网络规划和场地、环境等要求之外,还要防止系统信息在空间的扩散。计算机系统通过电磁辐射使信息被截获而失密的案例已经很多,在理论和技术支持下的验证工作也证实这种截取距离在几百甚至可达千米的复原显示技术给计算机系统信息的保密工作带来了极大的危害。为了防止系统中的信息在空间上的扩散,通常是在物理上采取一定的防护措施,来减少或干扰扩散出去的空间信号。正常的防范措施主要在三个方面:对主机房及重要信息存储、收发部门进行屏蔽处理,即建设一个具有高效屏蔽效能的屏蔽室,用它来安装运行主要设备,以防止磁鼓、磁带与高辐射设备等的信号外泄。为提
如何设置自动关机 比如你的电脑要在22:00关机，可以选择“开始→运行”，输入“at 22:00 Shutdown -s”， 这样，到了22点电脑就会出现“系统关机”对话框，默认有30秒钟的倒计时并提示你保存工 作。如果你想以倒计时的方式关机，可以输入“Shutdown.exe -s -t 3600”，这里表示60 分钟后自动关机，“3600”代表60分钟。 设置好自动关机后，如果想取消的话，可以在运行中输入“shutdown -a”。另外输入“shutdown -i”，则可以打开设置自动关机对话框，对自动关机进行设置。 Shutdown.exe的参数，每个都具有特定的用途，执行每一个都会产生不同的效果，比如“-s”就表示关闭本地计算机，“-a”表示取消关机操作，下面列出了更多参数，大家可以在Shutdown.exe中按需使用。 -f：强行关闭应用程序 -m \\计算机名：控制远程计算机 -i：显示图形用户界面，但必须是Shutdown的第一个选项 -l：注销当前用户 -r：关机并重启 -t 时间：设置关机倒计时 -c "消息内容"：输入关机对话框中的消息内容（不能超127个字符） 有时候，我们需要定时关闭计算机，下面介绍一个在Windows XP下实现定时关机的简单方 法。 指定系统在22分钟后自动关闭：点击“开始→运行”，在“打开”中输入命令“Shutdown －s －t 1320”（注意：引号不输入，参数之间有空格，1320的单位是秒），单击“确定” 按钮即可。 指定系统在某个时间（比如12：00）自动关闭：在“打开”中输入命令 “at 12：00 Shutdown －s”即可。
1.自动定时开机 此功能需要在BIOS中设定,并且你的主板的高级电源管理设置必须支持这个功能,不过现在大多数的主板全支持,放心使用吧. 步骤一: 启动系统按DEL,进入BIOS,然后选中"POWER MANAGEMENT FEATURES" (电源管理设置),回车进入.在电源窗口中找到"RESTORE ON AC/POWER LOSS"项,按回车,然后通过方向键设置其值为"POWER ON". 步骤二: 选中"RESUME ON RTC ALARM"并按回车,接着通过方向键设置其值为"ENABLE" 步骤三: 选中"RTC ALARM DATE"按回车,然后将其值设定为"EVERYDAY",表示每天都进行一样的操作,按照同样的方法,分别将"RTC ALARM HOURS""RTC ALARM MINUTE""RTM ALARM SECOND"设置为"07""15""00",这表示开机时间为早上的"7:15:00",全部设置好后,按F10,保存并重新启动. (由于不同的主板的BIOS不一样,有些主板的设置可能与上面的有一点点不同，但只要记住在电源管理中找到"WAKE UP"字样的就行了，因为这是设置定时开机的选项) 2自动登陆系统 电脑开机了，但是我们有些人设置了开机密码,不能自动登陆,怎么办? 我们可以通过修改注册表来实现自动登陆. 步骤一: 在运行中输入"REGEDIT",打开 HKEY_LOCAL_MACHINE\SOFTWARE\MICROSOFT\WINDOWSNT\CURRE NTVERSION\WINLOGON. 步骤二: 在右方的窗口中找到"DEFAULTUSERNAME"子键,将数值数据改为用户登陆帐户名,如果不存在该键值,就可以在右边的窗口中右键新建字符串,接着将"新值 #1"重命名为"DefaultUserName"后进行修改. 步骤三: 在右侧窗口中找到DefaultPassword,如果不存在,可以新建,然后将其键值设定为你的密码 步骤四: 按照以上的方法,我们在右边的窗口中新建一个名为"AutoAdminLogon"的键值,将其值设为1,然后退出重新启动就可以实现自动登陆 二:计划任务帮助我自动打开音乐 1让音乐吵醒你的耳朵 如果希望早上开机后可以听到美丽的音乐,可以按照下面的步骤进行, 步骤一:
xx校园网网络安全解决方案1 xx校园网网络安全解决方案 校园网网络是一个分层次的拓扑结构，因此网络的安全防护也需采用分层次的拓扑防护措施。即一个完整的校园网网络信息安全解决方案应该覆盖网络的各个层次，并且与安全管理相结合。 一、网络信息安全系统设计原则 1.1满足Internet分级管理需求 1.2需求、风险、代价平衡的原则 1.3综合性、整体性原则 1.4可用性原则 1.5分步实施原则 目前，对于新建网络及已投入运行的网络，必须尽快解决网络的安全保密问题，设计时应遵循如下思想： （1）大幅度地提高系统的安全性和保密性； （2）保持网络原有的性能特点，即对网络的协议和传输具有很好的透明性； （3）易于操作、维护，并便于自动化管理，而不增加或少增加附加操作；
（4）尽量不影响原网络拓扑结构，便于系统及系统功能的扩展； （5）安全保密系统具有较好的性能价格比，一次性投资，可以长期使用； （6）安全与密码产品具有合法性，并便于安全管理单位与密码管理单位的检查与监督。 基于上述思想，网络信息安全系统应遵循如下设计原则： 满足因特网的分级管理需求根据Internet网络规模大、用户众多的特点，对Internet/Intranet信息安全实施分级管理的解决方案，将对它的控制点分为三级实施安全管理。 --第一级：中心级网络，主要实现内外网隔离；内外网用户的访问控制；内部网的监控；内部网传输数据的备份与稽查。 --第二级：部门级，主要实现内部网与外部网用户的访问控制；同级部门间的访问控制；部门网内部的安全审计。 --第三级：终端/个人用户级，实现部门网内部主机的访问控制；数据库及终端信息资源的安全保护。 需求、风险、代价平衡的原则对任一网络，绝对安全难以达到，也不一定是必要的。对一个网络进行实际额研究（包括任务、性能、结构、可靠性、可维护性等），并对网络面临的威胁及可能承担的风险进行定性与定量相结合的分析，然后制定规范和措施，确定本系统的安全策略。
如果你想出门，但电脑又正在进行工作，这时就要用到自动关机。大多数 实现自动关机的 方法都是使用一些第三方软件，这样不仅麻烦，而且为实现这个小功能而 专门动用一个软件，显的小题大做了！其实Windows XP（Windows 2000也可以）自身就具备定时关机的功能，下面我们就来看看如何实现Windows XP的自动关机。 Windows XP的关机是由Shutdown.exe程序来控制的，位于 Windows\System32文件夹中。如果想让Windows 2000也实现同样的效果，可 以把Shutdown.exe复制到系统目录下。 比如你的电脑要在22:00关机，可以选择“开始→运行”，输入“at 22:00 Shutdown -s”，这样，到了22点电脑就会出现“系统关机”对话框，默认有30秒钟的倒计时并提示你保存工 作。如果你想以倒计时的方式关机，可以输入“Shutdown.exe -s -t 3600”，这 代表60分钟。 里表示60分钟后自动关机，“3600” 设置好自动关机后，如果想取消的话，可以在运行中输入“shutdown -a”。另外输入“shutdown -i”，则可以打开设置自动关机对话框，对自动关机进行设 置。 Shutdown.exe的参数，每个都具有特定的用途，执行每一个都会产生不同 的效果，比如“-s”就表示关闭本地计算机，“-a”表示取消关机操作，下面列出了 更多参数，大家可以在Shutdown.exe中按需使用。 -f： 强行关闭应用程序 计算机名： 控制远程计算机
-i： 显示图形用户界面，但必须是Shutdown的第一个选项 -l： 注销当前用户 -r： 关机并重启 -t时间： 设置关机倒计时 -c "消息内容"： 输入关机对话框中的消息内容（不能超127个字符） 有时候，我们需要定时关闭计算机，下面介绍一个在Windows XP下实现定时关机的简单方法。 指定系统在22分钟后自动关闭： 点击“开始→运行”，在“打开”中输入命令“Shutdown －s －t 1320”（注意： 引号不输入，参数之间有空格，1320的单位是秒），单击“确定”按钮即可。 指定系统在某个时间（比如12：00）自动关闭： 在“打开”中输入命令 “at 12：00 Shutdown－s”即可。 取消定时关机： 在“打开”中输入命令“Shutdown －a”即可。能的实现都是使用“Shutdown”命 。笔者认为最有令实现的，shutdown的参数是在命令行窗口中输入“shutdown”
校园网络安全设计方案 一、安全需求 1.1.1网络现状 随着信息技术的不断发展和网络信息的海量增加，校园网的安全形势日益严峻，目前校园网安全防护体系还存在一些问题，主要体现在：网络的安全防御能力较低，受到病毒、黑客的影响较大，对移动存储介质的上网监测手段不足，缺少综合、高效的网络安全防护和监控手段，削弱了网络应用的可靠性。因此，急需建立一套多层次的安全管理体系，加强校园网的安全防护和监控能力，为校园信息化建设奠定更加良好的网络安全基础。 经调查，现有校园网络拓扑图如下： 1.1.2应用和信息点
1.2.现有安全技术 1．操作系统和应用软件自身的身份认证功能，实现访问限制。 2．定期对重要数据进行备份数据备份。 3．每台校园网电脑安装有防毒杀毒软件。 1.3.安全需求 1．构建涵盖校园网所有入网设备的病毒立体防御体系。 计算机终端防病毒软件能及时有效地发现、抵御病毒的攻击和彻底清除病毒，通过计算机终端防病毒软件实现统一的安装、统一的管理和病毒库的更新。 2. 建立全天候监控的网络信息入侵检测体系 在校园网关键部位安装网络入侵检测系统，实时对网络和信息系统访问的异常行为进行监测和报警。 3. 建立高效可靠的内网安全管理体系 只有解决网络内部的安全问题，才可以排除网络中最大的安全隐患，内网安全管理体系可以从技术层面帮助网管人员处理好繁杂的客户端问题。 4. 建立虚拟专用网(VPN)和专用通道 使用VPN网关设备和相关技术手段，对机密性要求较高的用户建立虚拟专用网。 二．安全设计 1．1设计原则 根据防范安全攻击的安全需求、需要达到的安全目标、对应安全机制所需的安全服务等因素，参照SSE-CMM("系统安全工程能力成熟模型")和ISO17799(信息安全管理标准)等国际标准，综合考虑可实施性、可管理性、可扩展性、综合完备
Windows XP下如何设置电脑定时关机 我们在使用电脑经时，有时候会遇到一些紧急情况，需要暂时离开电脑，可是我们的电脑正在执行任务。比如下载资料、一些视频文件格式转换等等!我们知道电脑关闭还需要多长时间，也就是这些任务执行完毕还需要多长的时间!不想呆在电脑旁边看着它!我们就可以利用一下windows 的一些简单的命令来定时关机!那么如何实现电脑自动关机呢？下面我们就来学习下设置电脑定时关机的方法和步骤。 一、自动关机 其实现在我们只要在网上一搜，自动关机的小软件可以搜一大堆。也就是说大多数实现自动关机的方法都是使用一些第三方软件，这样不仅麻烦，而且为实现这个小功能而专门动用一个软件，显的小题大做了!其实Windows XP自身就具备定时关机的功能，下面我们就来看看如何实现Windows XP的自动关机。为找到答案，我们有必要先了解一下 WINDOWS XP的关机。Windows XP系统通过一个名为Shutdown.exe的程序来完成关机操作(位置在WindowsSystem下)，关机的时候调用shutdown.exe。因此，要实现自动关机，就是对shutdown.exe的调用。比如你的电脑要在12:00关机，可以选择―开始→运行‖，输入―at 12:00 Shutdown -s"，这样，到了12点电脑就会出现―系统关机‖对话框，默认有30秒钟的倒计时并提示你保存工作。如果你想以倒计时的方式关机，可以输入―Shutdown.exe -s -t 3600"，这里表示60分钟后自动关机，―3600"代表60分钟。结果如图: 系统关机 设置好自动关机后，如果想取消的话，可以在运行中输入―shutdown -a"。另外输入―shutdown -i"，则可以打开设置自动关机对话框，对自动关机进行设置。 Shutdown.exe的参数，每个都具有特定的用途，执行每一个都会产生不同的效果，比如―-s"就表示关闭本地计算机，―-a"表示取消关机操作，下面列出了更多参数，大家可以在Shutdown.exe中按需使用。 shutdown.exe -a 取消关机
电脑怎么设置自动关机及取消自动关机 电脑设置自动关机及取消自动关机方法 步骤一：首先进入win7开始运行对话框，可以按win+r组合快键键快速打开win7开始运行对框框。打开win7开始运行对话框后，我们再运行对话框的文本输入框里输入命令符:cmd 如下图: 步骤二：如上图，输入cmd命令符后，我们点击确认或按回车键即可进入windows cmd命令操作对话框，如下图： 步骤三：如上图，然后我们再在光标下输入shutdown –f –s –t 及数字，然后回车确定，注意了，后面的数字是以秒为单位的，比如输入54000，就代表在54000秒后win7系统将关闭计算机，大家可以根据实际工作要求设置多长时间后关机。当我们设定好后，win7桌面右下角会出现相应的关机提示，是不是很形象呢，如下图： 步骤四：通过以上几步简单的设置，我们即可轻松的设置win7系统在指定一段时间之后自动关机了，但如果有的时候，我们想再多玩下电脑，要取消自动关机该如何操作呢?方法依然简单，按照同样的方法，进入cmd命令框，再次驶入取消关机命令即可，取消命令为：shutdown –a 之后按回车键确认即可，如下图：通过上面的介绍，相信大家对win7自动关机设置已经都很清楚了，其实这里主要是运行的一些关机命令来实现，下面我们简单介绍下相关命令字符代表的含义：
-s 关闭此计算机 -f 强制运行的应用程序关闭而没有警告 -t xx 设置关闭的时间为xx秒后 -a 放弃系统关机 自动关机其实在我们生活中比较实用，比如晚上需要下载一些电影，我们又不能一直守着，这个时候可以让电脑在哪下载东东，我们估计号好下载所需要的时间，然后设置下自动关机，之后我们就可以去做别的事或者睡觉了。
网络安全课程设计 目录 一、校园网概况 二、校园网安全需求分析 三、产品选型和网络拓扑图介绍 四、操作系统安全配置与测试 五、应用服务器（WWW）安全配置 六、防病毒体系设计 七、防火墙设计、配置与测试 一、校园网概况
该校园网始建于2000年8月，至今已经历了四个主要发展阶段，网络覆盖已遍及现有的教学办公区和学生宿舍区。截止目前，校园网光缆铺设约一万二千米，信息点铺设接近一万，开设上网帐号8000多个，办理学校免费邮箱2000左右。 校园网主干现为双千兆环网结构。校园网接入均为千兆光纤到大楼，百兆交换到桌面，具有良好的网络性能。 校园网现有三条宽带出口并行接入Internet，500兆中国电信、100兆中国网通和100兆中国教育科研网，通过合理的路由策略，为校园网用户提供了良好的出口带宽。 校园网资源建设成效显著，现有资源服务包括大学门户网站、新闻网站、各学院和职能部门网站、安农科技网站、邮件服务、电子校务、毕博辅助教学平台、在线电视、VOD点播、音乐欣赏、公用FTP、文档下载、软件下载、知识园地、站点导航、在线帮助、系统补丁、网络安全、个人主页、计费服务、VPN、DHCP、域名服务等。还有外语学习平台，图书馆丰富的电子图书资源，教务处的学分制教学信息服务网、科技处的科研管理平台等。众多的资源服务构成了校园网的资源子网，为广大师生提供了良好的资源服务。 二、校园网安全需求分析 将安全策略、硬件及软件等方法结合起来，构成一个统一的防御系统，有效阻止非法用户进入网络，减少网络的安全风险。 定期进行漏洞扫描，审计跟踪，及时发现问题，解决问题。 通过入侵检测等方式实现实时安全监控，提供快速响应故障的手段，同时具备很好的安全取证措施。 使网络管理者能够很快重新组织被破坏了的文件或应用。使系统重新恢复到破坏前的状态，最大限度地减少损失。
电脑设置自动关机 方法一： 自动关机命令在开始→运行”，输入“at 22:00 Shutdown -s”就会自动关机！但是这个命令需要每天都进行设置。 如果不想每天都设置的话，可以加上/every参数，每周的1234567运行，就代表了每天了（eg：at 22:00 /every:M,T,W,Th,F,S,Su Shutdown -s -t 0） 取消命令是at /delete 方法二： 1、先在桌面上新建一个记事本,然后打开； 2、打开后输入内容shutdown -s -t 3600,代表一个小时后关机； 3、点击菜单,文件->另存为。 先把文件类型改为所有文件,取 文件名的时候,最后面要加[.bat], 即保存为批处理文件,以后只要 双击打开这个文件,就自动执行 1小时后自动关机。
方法三： Windows XP的关机是由Shutdown.exe程序来控制的，位于Windows\System32文件夹中。如 果想让Windows 2000也实现同样的效果，可以把Shutdown.exe复制到系统目录下。 比如你的电脑要在22:00关机，可以选择“开始→运行”，输入“at 22:00 Shutdown -s”， 这样，到了22点电脑就会出现“系统关机”对话框，默认有30秒钟的倒计时并提示你保存工 作。如果你想以倒计时的方式关机，可以输入“Shutdown.exe -s -t 3600”，这里表示60 分钟后自动关机，“3600”代表60分钟。 设置好自动关机后，如果想取消的话，可以在运行中输入“shutdown -a”。另外输入“shutdown -i”，则可以打开设置自动关机对话框，对自动关机进行设置。 Shutdown.exe的参数，每个都具有特定的用途，执行每一个都会产生不同的效果，比如“-s”就表示关闭本地计算机，“-a”表示取消关机操作，下面列出了更多参数，大家 可以在Shutdown.exe中按需使用。 -f：强行关闭应用程序 -m \\计算机名：控制远程计算机 -i：显示图形用户界面，但必须是Shutdown的第一个选项 -l：注销当前用户 -r：关机并重启 -t 时间：设置关机倒计时 -c "消息内容"：输入关机对话框中的消息内容（不能超127个字符） 有时候，我们需要定时关闭计算机，下面介绍一个在Windows XP下实现定时关机的简单方法。 指定系统在22分钟后自动关闭：点击“开始→运行”，在“打开”中输入命令“Shutdown －s －t 1320”（注意：引号不输入，参数之间有空格，1320的单位是秒），单击“确定”按钮即可。 指定系统在某个时间（比如12：00）自动关闭：在“打开”中输入命令 “at 12：00 Shutdown －s”即可。 取消定时关机：在“打开”中输入命令“Shutdown －a”即可。
自动关机命令设置 以晚上23：00关机为例，具体时间可以自己改。注意命令里的空格，复制时不要复制双引号。 1、开始/运行/输入“at 23:00 shutdown -s”将定于23：00自动关机。 中途取消自动关机，开始/运行/输入：shutdown -a。 2、用记事本输入： @echo off shutdown -s -t 0 保存为1.bat 然后拉入开始/程序/启动项。 或者重启： @echo off shutdown -r -t 5 保存为批处理文件,通过计划任务实现（省略）。 3、定时重启。用记事本输入：（定时在平时是看不到的，但在关机前30秒会提示，也可以加–t xx,自已设定） at 23:00 Shutdown -r 保存为bat格式，拉进开始/程序/启动。 取消自动关机直接在启动项里删除。 4、倒计时重启： Shutdown.exe -r -t 3600 倒计3600秒=60分钟重启。 5、打开c:\windows\sytem32/shutdown.exe，输入框中的 "shutdown"后面加上一个空格和"-r"，即"shutdown -r"，自己 设定时间就行了。 6、每天定时自动关机命令： at 23:00 /every:M,T,W,Th,F,S,Su 或： at 23:00 /every:M,T,W,Th,F,S,Su shutdown -s -f -t 120 -f 强制结束未响应程序 -t 120 延迟120秒 M,T,W,Th,F,S,Su 可选择星期几 at 23:00 /every:M,T,W,Th,F,S,Su shutdown -s -t 60 -c "早睡早起好身体"