当前位置：文档之家› COMBINING COMPETITIVE LEARNING NETWORKS OF VARIOUS REPRESENTATIONS FOR SEQUENTIAL DATA CLUS

COMBINING COMPETITIVE LEARNING NETWORKS OF VARIOUS REPRESENTATIONS FOR SEQUENTIAL DATA CLUS

Chapter 13

COMBINING COMPETITIVE LEARNING NETWORKS OF VARIOUS REPRESENTATIONS FOR SEQUENTIAL DATA CLUSTERING

Yun Yang and Ke Chen

School of Informatics

The University of Manchester

Manchester M60 1QD, U.K.

Yun.Yang@https://www.doczj.com/doc/997796585.html,, Ke.Chen@https://www.doczj.com/doc/997796585.html,

Abstract Sequential data clustering provides useful techniques for condensing and summarizing information conveyed in sequential data, which is demanded in

various fields ranging from time series analysis to video clip understanding. In

this chapter, we propose a novel approach to sequential data clustering by

combining multiple competitive learning networks incorporated by various

representations of sequential data and thus the clustering will be performed in

the feature space. In our approach, competitive learning networks of a rival-

penalized learning mechanism are employed for clustering analyses based on

different sequential data representations individually while an optimal

selection function is applied to find out a final consensus partition from

multiple partition candidates yielded by applying alternative consensus

functions to results of competitive learning on various representations. Thanks

to its capability of the rival penalized learning rules in automatic model

selection and the synergy of diverse partitions on various representations

resulting from diversified initialization and stopping conditions, our ensemble

learning approach yields favorite results especially in model selection, i.e. no

assumption on the number of clusters underlying a given data set is needed

prior to clustering analysis, which has been demonstrated in synthetic time

series and motion trajectory clustering analysis tasks.

Key words: Sequential data clustering, unsupervised ensemble learning, rival penalized competitive learning, local and global representations, model selection, motion

trajectory analysis, time series classification

1.INTRODUCTION

Sequential data are ubiquitous in the real world and there are many application areas ranging from multimedia information processing to financial data analysis. Unlike static data, there is a high amount of dependency among sequential data and the proper treatment of data dependency or correlation becomes critical in sequential data processing.

Clustering analysis provides an effective way to condensing and summarizing information conveyed in data, which is demanded by a number of application areas for organizing or discovering structures in data. The objective of clustering analysis is to partition a set of unlabeled objects into groups or clusters where all the objects grouped in the same cluster should be coherent or homogeneous. There are two core problems in clustering analysis; i.e., model selection and proper grouping. The former is seeking a solution that estimates the intrinsic number of clusters underlying a data set, while the latter demands a rule to group coherent objects together to form a cluster. From the perspective of machine learning, clustering analysis is an extremely difficult unsupervised learning task since it is inherently an ill-posed problem and its solution often violates some common assumptions [1]. There have been many studies in clustering analysis, which leads to various clustering algorithms categorized as either hierarchical or non-hierarchical algorithms [2]. However, the recent empirical studies in sequential data analysis reveal that most of the existing clustering algorithms do not work well for sequential data due to their special structure and data dependency [3], which presents a big challenge in clustering sequential data of a high dimensionality, very high feature correlation and a substantial amount of noise.

Competitive learning has been studied in the neural network community and turned out to be a useful tool for clustering analysis [4]. Among many competitive learning algorithms, however, few of them are capable of model selection and the number of clusters needs to be specified prior to clustering analysis. A rival penalized competitive learning (RPCL) algorithm was proposed to tackle both model selection and grouping problems under the framework of competitive learning [5]. More recently, one variant of RPCL, named rival penalization controlled competitive learning (RPCCL), has been proposed to improve its performance by using a data-driven de-learning rate [6]. Although RPCL and its variants have been successfully applied in static data clustering analysis [5],[6], our empirical studies indicate that the direct use of a RPCL-style algorithm in sequential data clustering tasks often fails to yield the satisfactory performance.

It is well known that the direct sequential data clustering often suffers from the curse-of-dimensionality and difficulties in capturing the long-term

Combining Competitive Learning Networks for Sequence Clustering 3 temporal dependency. Feature extraction is a process that distills the salient features from data and widely applied in pattern recognition. Unlike the direct use of sequential data, a feature extraction method often results in a parsimonious yet more effective representation of sequential data so that sequential data analysis can be fulfilled in the representation or feature space of low dimensionality. It has been proved that clustering analysis on the representation space often outperforms that on the original data space [2],[3].

A number of feature extraction methods for sequential data have been come up with from different perspectives [3], and they are roughly divided into two categories: local and global representations. A local representation tends to encode the local features precisely but is very difficult to characterize the global landscape of sequential data while a global representation models the landscape well by scarifying those local fine details in sequential data. To our knowledge, there is no universal representation that perfectly characterizes miscellaneous sequential data. As suggested by our recent work in non-verbal speech information processing [7], it is more likely that different representations need to be employed simultaneously to characterize complex sequential data entirely. Our earlier work in the real world sequential data classification, e.g. speaker recognition [8],[9], showed that the simultaneous use of different representations yields the significantly better performance than the use of an individual representation.

In this chapter, we propose a novel approach to sequential data clustering by an ensemble of RPCCL networks with different representations. We anticipate that the use of a RPCCL network, to a great extent, tackles the model selection problem without the use of prior information on the data domain while the ensemble of RPCCL networks tends to incorporate different representations to reach the synergy for clustering analysis. Recent researches in clustering ensembles [10],[11] provide feasible techniques to enable us to construct an ensemble RPCCL networks trained on different representations. We have applied our approach into time series and motion trajectory clustering tasks. Simulation results indicate that our approach yields the favorite performance in sequential data clustering. Our study reveals that the use of clustering ensemble techniques [10],[11] further improves the model selection performance, in particular, as individual RPCCL networks fail to estimate “right” cluster numbers.

The rest of this chapter is organized as follows. Sect. 2 describes three sequential data representations used in our simulations. Sec. 3 presents an ensemble competitive learning approach for sequential data clustering with the use of different representations and overviews the rival penalized competitive network. Sect. 4 reports simulation results in the time series and the motion trajectory clustering tasks. The last section draws conclusions.

2. SEQUENTIAL DATA REPRESENTATIONS

There have been a number of sequential data representations used in different application areas. In general, such representations can be classified into two categories: global and piecewise representations. A global representation is derived by modeling the sequential data via a set of basis functions and therefore coefficients in the parameter space forms a global representation that can be used to reconstruct the sequential data approximately. Some commonly used global representations are polynomial/spline curve fitting [12],[13], discrete Fourier transforms [14], discrete wavelet transforms [13]. In contrast, a piecewise representation is generated by partitioning the sequential data into segments at critical points based on a criterion then each segment will be characterized by a concise representation. As a result, all segment representations constitute an entire piecewise representation collectively, e.g. adaptive piecewise constant approximation [16] and curvature-based PCA segments [17]. As pointed out in Sect. 1, there is no universal representation that perfectly characterizes all sequential data. Thus, a global representation is often good at characterizing global features by smoothing out those local or fine details whilst a piecewise representation characterizes the local features very well but may fail to highlight the global landscape underlying sequential data.

Apparently, the complementary nature of global and piecewise representations suggests that it is likely to reach the synergy if we jointly use such representations to characterize sequential data. From the computation perspective, a representation of sequential data in a fixed dimension converts a temporal data clustering task into a static data clustering in the feature space. A global representation always leads to a feature vector of the fixed dimension regardless of the length of sequential data, while a piecewise representation often forms a feature vector of a dynamic dimension that depends on the nature of data, i.e. the number of critical points. Thus, most of the existing piecewise representations are not applicable in competitive learning due to dynamic dimensionalities of their feature vectors for different sequential data. As a result, we develop a coarse piecewise representation named piecewise local statistics (PLS) for our purpose. Thus we would adopt both the proposed piecewise representation and two typical global representations, i.e. polynomial curve fitting (PCF) and discrete Fourier transforms (DFT), in our simulations in order to demonstrate the benefit of using different representations for sequential data clustering analysis.

Most of sequential data can be regarded as time series of T sequential points expressed as T

t t x 1)}({=. For instance, motion trajectory resulting from

motion tracking has been employed to express video sequences. A motion

Combining Competitive Learning Networks for Sequence Clustering

trajectory is a 2-D spatiotemporal data of the notation T

t t y t x 1))}(),({(=, where

))(),((t y t x is the coordinates of an object tracked at frame t , and therefore can

also be treated as two separate time series T

t t x 1)}({=and T t t y 1)}({= by considering its x - and y -projection respectively. As a result, the representation of a

motion trajectory is simply a collective representation of two time series corresponding to its x - and y -projection. In the sequel, we shall present our PLS representation and briefly review two global representations only for univariate time series without a loss of generality.

2.1 Piecewise Local Statistics

Motivated by the short-term analysis in speech signal processing, we adopt a window based statistic analysis for time series. First of all, we use a window of the fixed size to block time series into a set of segments. For each segment, we estimate the 1st- and 2nd-order statistics used as features of this segment. For segment n , its local statistics, n μand n σ, are estimated by

])([||1 ,)(||1|

|||)1(12||||)1(1∑∑?+=?+=?==W n W n t n n W n W n t n t x W t x W μσμ (13.1) where |W| is the size of the window.

For time series, a PLS representation of a fixed dimension is formed by the collective notation of local statistic features of all segments though the estimate might be affected at the end of time series where the window is delimited.

2.2 Polynomial Curve Fitting

In [13], time series is modeled by fitting it to a parametric polynomial function

.)(0111αααα++???++=??t t t t x P P P P (13.2)

Here ),,1,0( P p p ???=αis the polynomial coefficient of the p th order. The fitting is carried out by minimizing a least-square error function by considering all sequential points of time series and the polynomial model of a given order, with respect to ),,1,0( P p p ???=α. All coefficients obtained via optimization constitute a PCF representation, a sequential point location dependent global representation of time series.

2.3 Discrete Fourier Transforms

Discrete Fourier transforms have been applied to derive a global representation of time series in frequency domain [14]. The DFT of time series T

t t x 1)}({= yields a set of Fourier coefficients:

1,,1,0 ),1

2exp()(1????=∑=?=T k T t T kt j t x T k a π (13.3) In order to form a robust representation in presence of noise, only few top K (K<

3. ENSEMBLE COMPETITIVE LEARNING MODEL In this section, we present an unsupervised ensemble learning model for combining multiple competitive learning networks individually trained on different representations for sequential data clustering. We first review the underpinning techniques used in our model, including the component RPCCL network [6] as well as clustering ensemble methods [10], and then describe our model. Finally we demonstrate how our unsupervised ensemble learning model and its components work with two 2-D synthetic data sets.

3.1 RPCCL Network

The RPCCL network [6] is a variant of the RPCL network [5] for competitive learning. Inheriting the strength of automatic model selection from the RPCL network, the RPCCL network overcomes the difficulty in determining de-learning rate and therefore is easier to use.

A RPCCL network consists of M binary units arranged in a layer. We use i u and i w to denote the output of unit i and its weight. All weights are initialized randomly at the beginning of learning. For a data set of N

objects,N n n 1}{=x , generating form M unknown intrinsic groups, the RPCL learning [5] tends to find out a set of proper weights adaptively so that

???==≤≤ otherwise, ,0||||min arg if ,1)(2

1j n j n i i u w x x (13.4) Here ||||?is the Euclidean norm.

Combining Competitive Learning Networks for Sequence Clustering

In order to obtain (13.4), the RPCL learning rule [5] has been developed, which consists of the following two steps:

1) Randomly choose an object n x from the data set N n n 1}{=x and for i=1,2,…,k , set the output of units by

??????==??===≠≤≤≤≤ otherwise

,0||||min arg , if ,1||||min arg , if ,1)(2,12

1j n j c j M j j n j M j n i r r i c c i u w x w x x ρρ (13.5) where ∑==M i i j j N N 1ρand j N is the total number of the winning occurrences of unit j so far.

2) Update the weights of units by

????=??=?=? otherwise ,01)( if ),)((1)( if ),(n i i n n r n i i n c i u u x w x x x w x w ηη (13.6)

and increment j N only if 1)(=n j u x or for the winner.

In Step 2, c η and )(n r x η are the learning and the de-learning rates. While the learning rate needs to be set prior to competitive learning, the data-driven de-learning rate is automatically determined by the RPCCL rule [6] as follows:

22||||}||||,||min{||)(c r c n c r c n r w w w x w w x ????=ηη (13.7) For competitive learning, the algorithm repeats Steps 1 and 2 until a pre-set stopping condition is reached.

In the RPCCL algorithm, there are several crucial factors that critically determine its performance of competitive learning. Our empirical studies in sequential data clustering indicate that its performance is sensitive to the learning rate selection as well as the initialization and the stopping conditions as exemplified in sec. 3.3. To our knowledge, there is no systematic way to choose the aforementioned parameters. Thus, our model cannot solely rely on this algorithm for robust clustering analysis; hence other advanced techniques need to be sought to ensure the robustness, which would be a reason why we employ clustering ensemble techniques in our model.

3.2 Clustering Ensemble Techniques

Clustering ensemble techniques have been recently studied to tackle several difficulties faced by an individual clustering algorithm. The basic idea underlying clustering ensemble techniques is combining multiple partitions of a data set by a consensus function to yield a final partition that more likely reflects the intrinsic structure of the data set.

A clustering ensemble method named Cluster Ensembles [10] has been recently proposed for the problem of combining multiple partitions of data without accessing the representations of data that determined these partitions, which results in a knowledge reusable framework. In [10], three consensus functions have been proposed and an objective function based on mutual information is further introduced to evaluate the performance of candidates yielded by different consensus functions optimally. However, all three consensus functions suffers from a weakness; i.e. the number of clusters in a final partition needs to be determined manually in advance or simply use the maximal number of clusters appearing in multiple partitions to be combined for model selection. Our empirical studies on different consensus functions indicate that the cluster-based similarity partitioning algorithm (CSPA) [10] often outperforms other two proposed in [10]. Therefore, we adopt the CSPA as one of consensus function in our ensemble learning model. Motivated by the clustering ensemble based on evidence accumulation [11], we also introduce an alternative consensus function that can automatically determine the number of clusters in the final partition. Below we briefly review the CSPA consensus function as well as the objective function for the final partition selection [10], and present the alternative consensus function.

3.2.1 Consensus functions

In the Cluster Ensemble [10], multiple partitions by an individual clustering algorithm are first mapped onto a hypergraph where its edge named hyperedge is allowed to connect any set of vertices. In the hypergraph, one vertex corresponds to one object in the data set and one cluster forms a hyperedge linked to all objects in the cluster. For partition q , a binary membership indicator matrix q H where a row corresponds to one object and a column refers to a binary encoding vector of one cluster in partition q . Thus concatenating all q H of multiple partitions leads to an adjacency matrix H by all objects in the data set versus all the available partitions.

Based on such a hypergraph representation, the CSPA specifies a consensus function as follows. The hypergraph representation encodes the

Combining Competitive Learning Networks for Sequence Clustering

piecewise similarity between any two objects; i.e. the similarity of one indicates two objects are grouped into the same cluster and a similarity of zero otherwise. Thus a similarity matrix S for all available partitions represented in a hypergraph is derived from the adjacency matrix H :

T P HH S ||1=where |P| is the number of partitions yielded by multiple-round clustering analyses. The average of similarities yielded from multiple partitions can be used to re-cluster objects to yield a final consensus.

We adapt the idea of [11] into a dendrogram-based similarity partitioning algorithm (DSPA) that determines the number of clusters in the final partition automatically. First of all, a co-associate matrix reflecting the relationship of all objects in multiple partitions is established where an element indicates the similarity defined by the number of occurrences as two specific objects are grouped into the same cluster. The co-associate matrix actually accumulates evidence and allows us to apply any clustering algorithm over this new similarity matrix for finding out a final partition. In our simulation, the average link method is applied to yield a dendrogram representation [2],[11]. Thus the proper number of clusters in the final partition is determined by cutting the dendrogram at the range of threshold points corresponding to the longest lifetime of clusters [2],[11].

3.2.2 Mutual-information based objective function

Although the aforementioned two consensus functions can be used individually to yield a clustering ensemble, their performance could be different as applied to data sets of various distributions. Without the prior information, it seems impossible to select a proper consensus function in advance to form a clustering ensemble. As a result, we apply a normalized mutual-information (NMI) based objective function proposed in [10] to measure the consistency between any two partitions:

∑∑∑∑====+=a b

b j

i a b b j a i ab ij

K i K j N b j N a i K i K j N N NN ab ij b a N N N P P 1111)log()log()

log(),(NMI (13.8)

Here a P and b P are labeling for two partitions that divide a data set of N

objects into a K and b K clusters, respectively. ab ij N is the number of shared objects between clusters a a i P C ∈and b b

j P C ∈, where there are a i N and b

j N objects in a i C and b j C . Based on (13.8), the optimal final partition can be determined by finding out the one that possesses maximal average mutual information with all |P | partitions available from multiple-round clustering analyses prior to the

clustering ensemble [10]. Thus finding the proper one from R various consensus functions can be done by

),(NMI max arg ||11*∑=≤≤=P p p r R r P P

P (13.9)

In other words, the consensus function yielding the partition *P is the proper one for the given data set.

3.2.3 Model Description

Based on the underpinning techniques presented above, we come up with an ensemble competitive learning model for sequential data clustering.

Figure 13.1. An ensemble competitive learning model with different representations.

As illustrated in Figure 13.1, our model consists of three modules; i.e. feature extraction, RPCCL competitive learning and clustering ensemble. In the feature extraction module, various representation methods of the complementary nature are demanded, as exemplified by three methods described in Sect. 2. Thus, sequential data are transformed into different representations to be the input of RPCCL networks. In the competitive learning module, a RPCCL network on an individual representation would

Combining Competitive Learning Networks for Sequence Clustering

be trained with its learning rules given in (13.5)-(13.7). Since the performance of a RPCCL network is often sensitive to the learning rate, the initialization and the stopping conditions, a RPCCL network would be trained in different conditions for several times, which yields multiple partitions. As a consequence, RPCCL networks on different representations lead to a collection of multiple partitions that can be encoded by a hypergraph described in Sect. 3.2.1. In the clustering ensemble module, two consensus functions presented in Sect. 3.2.1 are applied, respectively, by combining the collection of multiple partitions generated from the competitive learning to form the partition candidates from different perspectives. Finally the use of the objective function in (13.8) offers a mutual-information based selector in (13.9) that determines an optimal partition, named final partition, from a group of candidates for a given sequential data set.

3.3 Demonstrable Examples

In order to demonstrate the ensemble competitive learning and investigate the performance of a RPCCL network, we use a Gaussian mixture model to produce two 2-D data sets. In our experiment, we set the learning rate as 0.001 by default for the RPCCL network as suggested in [6], and choose six seed points whose initial positions are all randomly assigned in the input data space to test its model selection performance.

We produce 1000 data points randomly from a mixture of four Gaussian distributions as follows:

???????

???????????????????+????????????????????????+????????????????????????+????????????????????????=15.00015.0,11|32.015.00015.0,5.25.2|20.0 15.00015.0,5.21|22.015.00015.0,11|26.0)(x x x x x N N N N p (13.10) where each component ()Σm x ,|N denotes the Gaussian probability density function of variable x with the mean vector m and the covariance matrix Σ. As shown in Figure 13.2(a), a set of 2-D data points form four ball-shaped clusters where we mark the four clusters with different colors to indicate their ground truth; i.e. the data points marked by the same color are produced by the same Gaussian component distribution in (13.10). As a result, it is observed from Figure 13.2(a) that three clusters are overlapped moderately at the upper-right corner while one cluster is well separated from others at the lower-left corner. In terms of the shape and location of different data clusters, we would refer to this data set as a simple data set.

Figure 13.2. Performance on a simple data set. (a) The data set produced by (13.11). (b) The result of a RPCCL. (c) The result of a RPCCL ensemble.

Now we report results on the RPCCL on this data set for a single run. After 150 epochs, four seed points are moved into the appropriate position where is the center of each cluster, and the rest two seed points are pushed far away from the data points. The Figure 13.2(b) shows the clustering result by grouping the data points into the closed seed points. Next, we run the RPCCL with different initial positions of seed points for 20 trials, which produces 20 partitions. Then we use the ensemble technique presented in Sect. 3.2 to combine these partitions into a final single partition. As illustrated in Figure 13.2(c), the clustering result is exactly the same as the single RPCCL run.

From the results on the data set produced by (13.10), it is evident that the RPCCL network is capable of tackling the model selection problem by automatically determining the number of clusters without the use of any prior information on the data set. Moreover, the application of the ensemble learning on partitions produced by RPCCL network does not alter its correct partition and the number of clusters automatically determined previously. By manipulating parameters of a Gaussian mixture distribution, we could re-locate clusters in the data space by altering their mean vectors and shapes as well as spread levels of clusters by altering their covariance matrixes. In order to produce a more challenging data set of 1000 data sets, we employ another mixture of four Gaussian distributions as follows:

???

?????????????????????+????????????????????????+????????????????????????+????????????????????????=9.0059.0,340|25.09.0059.0,020|25.0 9009,010|25.09.0019.0,05|25.0)(x x x x x N N N N p (13.11) As illustrated in Figure 13.3(a), the 2-D data form four clusters marked by different colors and cluster shapes become irregular other than the ball shape shown in Figure 13.2(a). On the right hand side, two stripe-shaped

Combining Competitive Learning Networks for Sequence Clustering 13 clusters are well separated. On the left hand side, however, one dense cluster is wrapped by another sparse cluster.

Figure 13.3. Performance on a complex data set. (a) The data set produced by (13.11). (b) The result of a RPCCL. (c) The result of a RPCCL ensemble.

As shown in the Figure 13.3(b), the single RPCCL run results in the poor grouping for the data set produced by (13.11). From Figure 13.3(b), one strip-shaped cluster is wrongly partitioned into two separate clusters, and the wrapped dense cluster fails to be separate from the adjacent sparse cluster. In contrast, the application of the ensemble technique to the RPCCL networks yields a significantly improved partition as illustrated in Figure 13.3(c) where all the data points in the stripe-shaped cluster are correctly grouped and most of data points in the wrapped dense cluster are isolated from the data points in the adjacent sparse cluster though their boundary is less precise.

From this example, we would conclude that a single RPCCL run could fail to solve the model selection problem and fulfill the proper grouping for a data set like that produced by (13.11). Here we would refer to such a data set as a complex data set. Fortunately the ensemble learning technique appears to be helpful in this circumstance; to a great extent, it takes different partitions of RPCCL networks and manages to tackle both model selection and grouping problem for a complex data set.

In summary, the above examples demonstrate the effectiveness of combining multiple RPCCL networks for clustering.

4.SIMULATION

In this section, we present our experimental methodology and simulation results. Although the outcome of clustering analysis can be used for miscellaneous tasks, we focus on only the clustering-based classification

tasks in simulations. Clustering analysis is first done by partitioning a set of training data into several clusters. The training data are labeled based on clustering analysis and then used as prototypes to classify testing data unseen during training.

In our simulations, we apply an ensemble of RPCCL competitive learning networks to a synthetic time series benchmark, the Cylinder-Bell-Funnel data set [3], and an objective trajectory benchmark, the CAVIAR visual tracking database [18]. Below we report the performance of our model in two benchmark clustering tasks.

4.1 The Cylinder-Bell-Funnel Data Set

This data set has been used as a benchmark in sequential data mining [3]. The data are generated by three time series functions:

).()()()()6()(),

()()6()(),

()()6()(],[],[],[t a b b t t x t f t t x t b t t x t c b a b a b a εκεκεκ+??+=++=++= (13.12)

where κ and ε(t) are drawn from a Gaussian distribution N(0,1), a an b are two integers randomly drawn from intervals [16, 32] and [48, 128], and )(],[t x b a is defined as one if a t b ≤≤ and zero otherwise. Three stochastic functions in (13.12) randomly generate time series of 128 frames corresponding to three classes: Cylinder, Bell and Funnel. In our simulations, we generated 200 samples for each class and then randomly divided them into two subsets of equal sizes; one for training and the other for test.

Figure 13.4. Distribution of samples in various PCA representation manifolds formed by the 1st and 2nd principal components. (a) PCL. (b) PCF. (c) DFT.

Combining Competitive Learning Networks for Sequence Clustering

Feature extraction was first done for all training samples to generate three representations for each time series. In order to investigate the nature of different representations, we perform the principal component analysis (PCA) on three representations for the training set. As illustrated in Figure 13.4, different representations lead to diverse distributions of samples in the PCA representation subspace so that intra- and inter-class variations of samples can be various in different representation space. To some extent, this justifies that the use of multiple complementary representations paves a way to promote the diversity of ensemble clustering.

For robust competitive learning, the RPCCL networks were independently trained on individual representations for 20 times by setting various learning rates, initialization and stopping conditions so that 20 partitions can be generated for each representation. In order to evaluate the model selection performance, we deliberately choose the initial number of units in a RPCCL network as five although the ground truth is available, i.e. three classes underlying the data set. Thus the RPCCL network needs to estimate the proper number of clusters during its rival penalized learning. Once multiple partitions are available, two consensus functions are applied to yield final partition candidates and then the mutual-information based selector to determine the optimal final partition.

Table 13.1. Classification Rate (%) of Various Methods and Parameters.

For statistical reliability, the aforementioned experiment is repeated 40 times and the averaging results are reported here. Table 13.1 shows the performance of single RPCCL networks with single representations, an ensemble of RPCCL networks with single representations and an ensemble Representation Method Parameter in Representation Single Clustering (classification rate) training testing Clustering Ensemble (classification rate) training testing PLS |W | =3 |W |=6 |W |=12 73.0 71.9 73.1 72.4 70.6 69.1 78.9 77.9 79.4 78.1 78.5 75.1 PCF P =3 P =4 P =5 60.2 59.2 61.1 60.2 58.3 54.9 63.3 62.1 64.1 62.3

62.2 60.0 DFT K =8 K =16 K =32 66.7 63.3 75.7 72.9 60.7 58.1 67.9 64.6 76.9 74.7 63.2 60.7 Different Representations |W | = 6 P =4 K =16

N/A 82.7 81.0

of RPCCL networks with different representations. From Table 13.1, it is observed that the performance of RPCCL networks changes on single representations due to different distributions demonstrated in Figure 13.4 and ensembles of RPCCL networks, in particular, with different representations significantly improve the performance on both training and testing sets. In addition, the performance is relatively insensitive to parameters of different representations. Note the classification rate is calculated by comparing the results to the ground truth. Here we emphasize that all results are achieved based on initializing the RPCCL network with the number of clusters (i.e., five) inconsistent with the ground truth (i.e., three).

Using the proper number of clusters given in the ground truth, many clustering algorithms, e.g. K-means and SOM, with an individual representation, have been applied to the data set, and typical classification rates reported are about between 75% and 80% [3],[19]. For comparison, our ensemble model has also been used on the same condition, i.e. the use of ground truth information, and the averaging classification rates on training and testing sets are 85.2% and 83.7%, respectively. In contrast to the performance of our ensemble model in Table 13.1 without the use of prior information on the data set, we conclude that our ensemble competitive learning model yields robust clustering analysis and hence suitable for being applied in an unknown environment.

4.2The CAVIA Visual Tracking Database

The CAVIA database is a benchmark designed for video content analysis [18]. From the manually annotated video sequences of pedestrians, a set of 222 high-quality moving trajectories, as shown in Figure 13.6(a), are achieved for clustering analysis.

In our simulation, our model is first applied to all trajectories for clustering analysis. Since the information on the “right” number of clusters is unavailable, 20 units are initially adopted in the RPCCL network for competitive learning based on our belief that too many clusters make little sense for a higher level analysis. Similarly, the RPCCL network based on an individual representation yields 20 partitions by training repeatedly with different conditions and all 60 partitions are combined by different consensus functions to yield final partition candidates. The final partition is eventually chosen by the mutual-information based selector. As illustrated in Figure. 13.5, our model automatically partitions the 222 moving trajectories into 14 clusters. Human visual inspection suggests that similar motion trajectories have been grouped together properly while dissimilar ones are distributed into different clusters. For example, those trajectories

Combining Competitive Learning Networks for Sequence Clustering 17 corresponding to moving from left-to-right and right-to-left are properly grouped into two separate clusters as shown in Figure 13.5(f) and (g).

Figure 13.5. A clustering analysis of all moving trajectories in CAVIAR database by our ensemble competitive learning model. Plots in (a)-(n) correspond to 14 clusters of moving trajectories in the final partition.

The CAVIAR database has also been used in [19] to evaluate their method where the SOM with an individual representation was used. In their simulation, the number of clusters was determined manually, and all trajectories are simply grouped into nine clusters. Although most of clusters achieved in their method are consistent with ours, a couple of separate clusters achieved in our model, e.g. clusters shown in Figure 13.5(a) and (c), are merged into a single cluster in theirs. If trajectories in separate clusters express different events, such a merge would lead to a difficulty in a higher level analysis.

Figure 13.6. All moving trajectories with background scene in CAVIAR database and typical clustering of RPCCL network ensembles only with individual representations. (a) All moving trajectories. (b) PLS. (c) PCF. (d) DFT.

For demonstrating the benefit of jointly using different representations in our model, Figure 13.6(b)-(d) illustrate three typical clusters of moving trajectories yielded by RPCCL network ensembles only with individual representations. Visual inspection indicates inadequate clustering achieved due to their weakness of individual representations. The cluster shown in Figure 13.6(b) groups trajectories from two clusters shown in Figure 13.5(b) and 13.5(m) since the PLS representation highlights local features but relatively neglects global characteristics. In Figure 13.6(c), some “orthogonal” trajectories are improperly grouped together due to the limited representation capability of the PCF. Similarly, trajectories with the same orientation but starting from different positions are incorrectly grouped together because the DFT is a representation independent of spatial locations. In contrast to the clustering results shown in Figure 13.5, all of such less meaningful clusters no longer exist thanks to the joint use of different representations.

Combining Competitive Learning Networks for Sequence Clustering

For further evaluation, we have done two additional experiments in classification: a) adding different amounts of Gaussian noise,),0(σN , to a range of coordinates in the database; b) randomly removing different parts of each moving trajectory and producing its noisy missing data version by further adding the Gaussian noise with 1.0=σ. The former tends to generate testing data sets whilst the latter simulates common scenarios that a moving object tracked is occluded by other objects or the background so that a tracking algorithm has to produce a trajectory with missing data.

Table 13.2. Classification Rate (%) of Ensemble methods on CAVIAR.

Figure 13.7. Performance of our ensemble model with different representations on CAVIAR as missing data appear.

Representation Method Parameter in Representation (Clustering Ensemble) σ =0.1 σ =0.2 σ =0.3 σ =0.4 PLS |W |=150 95.8 91.7 87.2 84.3

PCF P =4 89.5 83.6 78.0 70.2

DFT K =16 93.9 88.2 87.2 77.4

Different Representations |W | = 150 P = 4

K = 16 96.4 92.3 87.8 84.7

Table 13.2 lists results for classification based on the clustering analysis in Figure 13.5. Apparently doing so highly relies on the quality of clustering analysis that was backed by the human visual inspection. The results in Table 13.2 reveal that the classification performance would be satisfactory and acceptable especially as a substantial amount of noise is added. Figure 13.7 illustrates the evolution of performance degradation caused by different amounts of missing data. It is apparent that our model is capable of dealing with trajectories of an occluded object tracked.

In summary, all above simulation results suggest that our model leads to a robust clustering analysis without the use of prior information on a given data set and, therefore, its outcome can be used for higher level video content analyses.

5.CONCLUSION

In this chapter, we have presented an unsupervised ensemble learning model for sequential data clustering by combining multiple competitive learning networks with different representations. Without the use of prior information on a given data set, simulation results on different types of sequential data demonstrate that our model yields favorite results in both model selection and grouping. In particular, the joint use of different representations leads to a significant improvement.

There are several issues to be studied in our ongoing research. First, three representations used in this chapter are coarse for representing sequential data, which is simple for a demonstration purpose. Exploration of effective yet complementary representations would be an urgent topic to be investigated in our research. Next, our simulations indicate that the use of the RPCCL network results in two-fold effects. On the one hand, its automatic model selection capability is helpful to cope with problems in an unknown environment. On the other hand, our simulation results including those not reported here due to space indicate that its learning rule seems to hinder generating truly diverse partitions; although the use of different learning rates, the initialization and the stopping conditions leads to different partitions, the correlation among them seems quite high. To our knowledge, there has been no theoretic analysis available so far regarding combining the highly correlated partitions. Nevertheless the theoretic analysis in combining classifiers suggests that combining highly correlated classifiers is unlikely to yield the considerate improvement in classification [20]. We trust that their conclusion should be more or less applicable to ensemble clustering. As a result, we need to investigate the correlation problem behind the RPCL style

知识管理和信息管理之间的联系和区别_[全文]

引言：明确知识管理和信息管理之间的关系对两者的理论研究和实际应用都有重要意义。从本质上看,知识管理既是在信息管理基础上的继承和进一步发展,也是对信息管理的扬弃,两者虽有一定区别,但更重要的是两者之间的联系。知识管理和信息管理之间的联系和区别 By AMT 宋亮相对知识管理而言,信息管理已有了较长的发展历史,尽管人们对信息管理的理解各异,定义多样,至今尚未取得共识,但由于信息对科学研究、管理和决策的重要作用,信息管理一词一度成为研究的热点。知识管理这一概念最近几年才提出,而今知识管理与创新已成为企业、国家和社会发展的核心,重要原因就在于知识已取代资本成为企业成长的根本动力和竞争力的主要源泉。信息管理与知识管理的关系如何?有人将知识管理与信息管理对立起来,强调将两者区分开;也有不少人认为知识管理就是信息管理;多数人则认为,信息管理与知识管理虽然存在较大的差异,但两者关系密切,正是因为知识和信息本身的密切关系。明确这两者的关系对两者的理论研究和实际应用都有重要意义。从本质上看,知识管理既是在信息管理基础上的继承和进一步发展,也是对信息管理的扬弃,两者虽有一定区别,但更重要的是两者之间的联系。一、信息管理与知识管理之概念 1、信息管理 “信息管理”这个术语自本世纪70年代在国外提出以来,使用频率越来越高。关于“信息管理”的概念,国外也存在多种不同的解释。1>.人们公认的信息管理概念可以总结如下:信息管理是实观组织目录、满足组织的要求,解决组织的环境问题而对信息资源进行开发、规划、控制、集成、利用的一种战略管理。同时认为信息管理的发展分为三个时期:以图书馆工作为基础的传统管理时期,以信息系统为特征的技术管理时期和以信息资源管理为特征的资源管理时期。 2、知识管理由于知识管理是管理领域的新生事物，所以目前还没有一个被大家广泛认可的定义。Karl E Sverby从认识论的角度对知识管理进行了定义，认为知识管理是“利用组织的无形资产创造价值的艺术”。日本东京一桥大学著名知识学教授野中郁次郎认为知识分为显性知识和隐性知识，显性知识是已经总结好的被基本接受的正式知识，以数字化形式存在或者可直接数字化，易于传播；隐性知识是尚未从员工头脑中总结出来或者未被基本接受的非正式知识，是基于直觉、主观认识、和信仰的经验性知识。显性知识比较容易共享，但是创新的根本来源是隐性知识。野中郁次郎在此基础上提出了知识创新的SECI模型:其中社会化(Socialization)，即通过共享经验产生新的意会性知识的过程；外化(Externalization)，即把隐性知识表达出来成为显性知识的过程；组合（Combination），即显性知识组合形成更复杂更系统的显性知识体系的过程；内化（Internalization），即把显性知识转变为隐性知识，成为企业的个人与团体的实际能力的过程。

新版人教版高中语文课本的目录。

必修一阅读鉴赏第一单元1.*沁园春?长沙……………………………………毛泽东3 2.诗两首雨巷…………………………………………戴望舒6 再别康桥………………………………………徐志摩8 3.大堰河--我的保姆………………………………艾青10 第二单元4.烛之武退秦师………………………………….《左传》16 5.荆轲刺秦王………………………………….《战国策》18 6.*鸿门宴……………………………………..司马迁22 第三单元7.记念刘和珍君……………………………………鲁迅27 8.小狗包弟……………………………………….巴金32 9.*记梁任公先生的一次演讲…………………………梁实秋36 第四单元10.短新闻两篇别了，“不列颠尼亚”…………………………周婷杨兴39 奥斯维辛没有什么新闻…………………………罗森塔尔41 11.包身工………………………………………..夏衍44 12.*飞向太空的航程……………………….贾永曹智白瑞雪52 必修二阅读鉴赏第一单元1.荷塘月色…………………………………..朱自清2.故都的秋…………………………………..郁达夫3.*囚绿记…………………………………..陆蠡第二单元4.《诗经》两首氓采薇5.离骚………………………………………屈原6.*《孔雀东南飞》(并序) 7.*诗三首涉江采芙蓉《古诗十九首》短歌行……………………………………曹操归园田居(其一)…………………………..陶渊明第三单元8.兰亭集序……………………………………王羲之9.赤壁赋……………………………………..苏轼10.*游褒禅山记………………………………王安石第四单元11.就任北京大学校长之演说……………………..蔡元培12.我有一个梦想………………………………马丁?路德?金1 3.*在马克思墓前的讲话…………………………恩格斯第三册阅读鉴赏第一单元1.林黛玉进贾府………………………………….曹雪芹2.祝福………………………………………..鲁迅3. *老人与海…………………………………….海明威第二单元4.蜀道难……………………………………….李白5.杜甫诗三首秋兴八首(其一) 咏怀古迹(其三) 登高6.琵琶行(并序)………………………………..白居易7.*李商隐诗两首锦瑟马嵬(其二) 第三单元8.寡人之于国也…………………………………《孟子》9.劝学……………………………………….《荀子》10.*过秦论…………………………………….贾谊11.*师说………………………………………韩愈第四单元12.动物游戏之谜………………………………..周立明13.宇宙的边疆………………………………….卡尔?萨根14.*凤蝶外传……………………………………董纯才15.*一名物理学家的教育历程……………………….加来道雄第四册阅读鉴赏第一单元1.窦娥冤………………………………………..关汉卿2.雷雨………………………………………….曹禹3.*哈姆莱特……………………………………莎士比亚第二单元4.柳永词两首望海潮(东南形胜) 雨霖铃(寒蝉凄切) 5.苏轼词两首念奴娇?赤壁怀古定风波(莫听穿林打叶声) 6.辛弃疾词两首水龙吟?登建康赏心亭永遇乐?京口北固亭怀古7.*李清照词两首醉花阴(薄雾浓云愁永昼) 声声慢(寻寻觅觅) 第三单元8.拿来主义……………………………………….鲁迅9.父母与孩子之间的爱……………………………..弗罗姆10.*短文三篇热爱生

信息管理与知识管理典型例题讲解

信息管理与知识管理典型例题讲解 1.（）就是从事信息分析，运用专门的专业或技术去解决问题、提出想法，以及创造新产品和服务的工作。 2.共享行为的产生与行为的效果取决于多种因素乃至人们的心理活动过程，（）可以解决分享行为中的许多心理因素及社会影响。 3.从总体来看，目前的信息管理正处（）阶段。 4.知识员工最主要的特点是（）。 5.从数量上来讲，是知识员工的主体，从职业发展角度来看，这是知识员工职业发展的起点的知识员工类型是（）。 6.人类在经历了原始经济、农业经济和工业经济三个阶段后，目前正进入第四个阶段——（），一种全新的经济形态时代。 1.对信息管理而言，( )是人类信息活动产生和发展的源动力，信息需求的不断变化和增长与社会信息稀缺的矛盾是信息管理产生的内在机制，这是信息管理的实质所在。 2.（）就是从事信息分析，运用专门的专业或技术去解决问题、提出想法，以及创造新产品和服务的工作。 3.下列不属于信息收集的特点（） 4.信息管理的核心技术是（） 5.下列不属于信息的空间状态（） 6.从网络信息服务的深度来看，网络信息服务可以分 1.知识管理与信息管理的关系的（）观点是指，知识管理在历史上曾被视为信息管理的一个阶段，近年来才从信息管理中孵化出来，成为一个新的管理领域。 2.著名的Delicious就是一项叫做社会化书签的网络服务。截至2008年8月，美味书签已经拥有（）万注册用户和160万的URL书签。 3.激励的最终目的是（）

4.经济的全球一体化是知识管理出现的（） 5.存储介质看，磁介质和光介质出现之后，信息的记录和传播越来越（） 6.下列不属于信息的作用的选项（） 1.社会的信息化和社会化信息网络的发展从几方面影响着各类用户的信息需求。（） 2.个性化信息服务的根本就是“（）”，根据不同用户的行为、兴趣、爱好和习惯，为用户搜索、组织、选择、推荐更具针对性的信息服务。 3.丰田的案例表明：加入( )的供应商确实能够更快地获取知识和信息，从而促动了生产水平和能力的提高。 4.不属于信息资源管理的形成与发展需具备三个条件（） 5.下列不属于文献性的信息是（） 6.下列不属于（）职业特点划分的

连锁咖啡厅顾客满意度涉入程度对忠诚度影响之研究以高雄市星巴克为例

连锁咖啡厅顾客满意度涉入程度对忠诚度影响之研究以高雄市星巴克为例 Document number【SA80SAB-SAA9SYT-SAATC-SA6UT-SA18】

连锁咖啡厅顾客满意度对顾客忠诚度之影响－以高雄市星巴克为例 The Effects of Customer Satisfaction on Customer Loyalty—An Empirical study of Starbucks Coffee Stores in Kaohsiung City 吴明峻 Ming-Chun Wu 高雄应用科技大学观光管理系四观二甲学号：06 中文摘要本研究主要在探讨连锁咖啡厅顾客满意度对顾客忠诚度的影响。本研究以高雄市为研究地区，并选择8间星巴克连锁咖啡厅的顾客作为研究对象，问卷至2006年1月底回收完毕。本研究将顾客满意度分为五类，分别是咖啡、餐点、服务、咖啡厅内的设施与气氛和企业形象与整体价值感；将顾客忠诚度分为三类，分别是顾客再购意愿、向他人推荐意愿和价格容忍度，并使用李克特量表来进行测量。根据过往研究预期得知以下研究结果： 1.人口统计变项与消费型态变项有关。 2.人口统计变项与消费型态变项跟顾客满意度有关。 3.人口统计变项与消费型态变项跟顾客忠诚度有关。 4.顾客满意度对顾客忠诚度相互影响。关键词：连锁、咖啡厅、顾客满意度、顾客忠诚度 E-mail

一、绪论研究动机近年来，国内咖啡消费人口迅速增加，国外知名咖啡连锁品牌相继进入台湾，全都是因为看好国内咖啡消费市场。在国内较知名的连锁咖啡厅像是星巴克、西雅图极品等。本研究针对连锁咖啡厅之顾客满意度与顾客忠诚度间关系加以探讨。研究目的本研究所要探讨的是顾客满意度对顾客忠诚度的影响，以国内知名的连锁咖啡厅星巴克之顾客为研究对象。本研究有下列五项研究目的： 1.以星巴克为例，探讨连锁咖啡厅的顾客满意度对顾客忠诚度之影响。 2.以星巴克为例，探讨顾客满意度与顾客忠诚度之间的关系。 3.探讨人口统计变项与消费型态变项是否有相关。 4.探讨人口统计变项与消费型态变项跟顾客满意度是否有相关。 5.探讨人口统计变项与消费型态变项跟顾客忠诚度是否有相关。二、文献回顾连锁咖啡厅经营风格分类根据行政院（1996）所颁布的「中华民国行业标准分类」，咖啡厅是属於九大行业中的商业类之饮食业。而国内咖啡厅由於创业背景、风格以及产品组合等方面有其独特的特质，使得经营型态与风格呈现多元化的风貌。依照中华民国连锁店协会（1999）对咖啡产业调查指出，台湾目前的咖啡厅可分成以下几类：

高中外研社英语选修六Module5课文Frankenstein's Monster

Frankenstein's Monster Part 1 The story of Frankenstein Frankenstein is a young scientist/ from Geneva, in Switzerland. While studying at university, he discovers the secret of how to give life/ to lifeless matter. Using bones from dead bodies, he creates a creature/ that resembles a human being/ and gives it life. The creature, which is unusually large/ and strong, is extremely ugly, and terrifies all those/ who see it. However, the monster, who has learnt to speak, is intelligent/ and has human emotions. Lonely and unhappy, he begins to hate his creator, Frankenstein. When Frankenstein refuses to create a wife/ for him, the monster murders Frankenstein's brother, his best friend Clerval, and finally, Frankenstein's wife Elizabeth. The scientist chases the creature/ to the Arctic/ in order to destroy him, but he dies there. At the end of the story, the monster disappears into the ice/ and snow/ to end his own life. Part 2 Extract from Frankenstein It was on a cold November night/ that I saw my creation/ for the first time. Feeling very anxious, I prepared the equipment/ that would give life/ to the thing/ that lay at my feet. It was already one/ in the morning/ and the rain/ fell against the window. My candle was almost burnt out when, by its tiny light，I saw the yellow eye of the creature open. It breathed hard, and moved its arms and legs. How can I describe my emotions/ when I saw this happen? How can I describe the monster who I had worked/ so hard/ to create? I had tried to make him beautiful. Beautiful! He was the ugliest thing/ I had ever seen! You could see the veins/ beneath his yellow skin. His hair was black/ and his teeth were white. But these things contrasted horribly with his yellow eyes, his wrinkled yellow skin and black lips. I had worked/ for nearly two years/ with one aim only, to give life to a lifeless body. For this/ I had not slept, I had destroyed my health. I had wanted it more than anything/ in the world. But now/ I had finished, the beauty of the dream vanished, and horror and disgust/ filled my heart. Now/ my only thoughts were, "I wish I had not created this creature, I wish I was on the other side of the world, I wish I could disappear!” When he turned to look at me, I felt unable to stay in the same room as him. I rushed out, and /for a long time/ I walked up and down my bedroom. At last/ I threw myself on the bed/ in my clothes, trying to find a few moments of sleep. But although I slept, I had terrible dreams. I dreamt I saw my fiancée/ walking in the streets of our town. She looked well/ and happy/ but as I kissed her lips，they became pale, as if she were dead. Her face changed and I thought/ I held the body of my dead mother/ in my arms. I woke, shaking with fear. At that same moment，I saw the creature/ that I had created. He was standing/by my bed/ and watching me. His

人教版高中语文必修必背课文精编WORD版

人教版高中语文必修必背课文精编W O R D版 IBM system office room 【A0816H-A0912AAAHH-GX8Q8-GNTHHJ8】

必修1 沁园春·长沙（全文）毛泽东独立寒秋，湘江北去，橘子洲头。看万山红遍，层林尽染，漫江碧透，百舸争流。鹰击长空，鱼翔浅底，万类霜天竞自由。怅寥廓，问苍茫大地，谁主沉浮。携来百侣曾游，忆往昔峥嵘岁月稠。

恰同学少年，风华正茂，书生意气，挥斥方遒。指点江山，激扬文字，粪土当年万户侯。曾记否，到中流击水，浪遏飞舟。雨巷（全文）戴望舒撑着油纸伞，独自彷徨在悠长、悠长又寂寥的雨巷，我希望逢着一个丁香一样地

结着愁怨的姑娘。她是有丁香一样的颜色，丁香一样的芬芳，丁香一样的忧愁，在雨中哀怨，哀怨又彷徨；她彷徨在这寂寥的雨巷，撑着油纸伞像我一样，像我一样地默默彳亍着冷漠、凄清，又惆怅。她默默地走近，走近，又投出太息一般的眼光

她飘过像梦一般地，像梦一般地凄婉迷茫。像梦中飘过一枝丁香地，我身旁飘过这个女郎；她默默地远了，远了，到了颓圮的篱墙，走尽这雨巷。在雨的哀曲里，消了她的颜色，散了她的芬芳，消散了，甚至她的太息般的眼光丁香般的惆怅。撑着油纸伞，独自

彷徨在悠长、悠长又寂寥的雨巷，我希望飘过一个丁香一样地结着愁怨的姑娘。再别康桥（全文）徐志摩轻轻的我走了，正如我轻轻的来；我轻轻的招手，作别西天的云彩。那河畔的金柳，是夕阳中的新娘；波光里的艳影，在我的心头荡漾。软泥上的青荇，油油的在水底招摇；在康河的柔波里，我甘心做一条水草！那榆荫下的一潭，不是清泉，是天上虹揉碎在浮藻间，沉淀着彩虹似的梦。寻梦？撑一支长篙，向青草更青处漫溯，满载一船星辉，在星辉斑斓里放歌。

(完整版)Unit7TheMonster课文翻译综合教程四

Unit 7 The Monster Deems Taylor 1He was an undersized little man, with a head too big for his body ― a sickly little man. His nerves were bad. He had skin trouble. It was agony for him to wear anything next to his skin coarser than silk. And he had delusions of grandeur. 2He was a monster of conceit. Never for one minute did he look at the world or at people, except in relation to himself. He believed himself to be one of the greatest dramatists in the world, one of the greatest thinkers, and one of the greatest composers. To hear him talk, he was Shakespeare, and Beethoven, and Plato, rolled into one. He was one of the most exhausting conversationalists that ever lived. Sometimes he was brilliant; sometimes he was maddeningly tiresome. But whether he was being brilliant or dull, he had one sole topic of conversation: himself. What he thought and what he did. 3He had a mania for being in the right. The slightest hint of disagreement, from anyone, on the most trivial point, was enough to set him off on a harangue that might last for hours, in which he proved himself right in so many ways, and with such exhausting volubility, that in the end his hearer, stunned and deafened, would agree with him, for the sake of peace. 4It never occurred to him that he and his doing were not of the most intense and fascinating interest to anyone with whom he came in contact. He had theories about almost any subject under the sun, including vegetarianism, the drama, politics, and music; and in support of these theories he wrote pamphlets, letters, books ... thousands upon thousands of words, hundreds and hundreds of pages. He not only wrote these things, and published them ― usually at somebody else’s expense ― but he would sit and read them aloud, for hours, to his friends, and his family. 5He had the emotional stability of a six-year-old child. When he felt out of sorts, he would rave and stamp, or sink into suicidal gloom and talk darkly of going to the East to end his days as a Buddhist monk. Ten minutes later, when something pleased him he would rush out of doors and run around the garden, or jump up and down off the sofa, or stand on his head. He could be grief-stricken over the death of a pet dog, and could be callous and heartless to a degree that would have made a Roman emperor shudder. 6He was almost innocent of any sense of responsibility. He was convinced that

(完整版)信息管理与知识管理

信息管理与知识管理（）是第一生产力,科学技术（）是物质系统运动的本质特征，是物质系统运动的方式、运动的状态和运动的有序性信息按信息的加工和集约程度分（）N次信息源不是阻碍企业技术创新的主要因素主要包括（）市场需求薄弱不属于 E.M.Trauth博士在20世纪70年代首次提出信息管理的概念，他将信息管理分内容为（）网络信息管理不属于W.Jones将个人信息分为三类（）个人思考的不属于信息的经济价值的特点（）永久性不属于信息的社会价值特征的实（）广泛性不属于信息资源管理的形成与发展需具备三个条件（）信息内容存储介质看，磁介质和光介质出现之后，信息的记录和传播越来越（）方便快捷国外信息管理诞生的源头主要有（）个1个即信息是否帮助个人获得更好的发展，指的是（）个人发展社会的信息化和社会化信息网络的发展从几方面影响着各类用户的信息需求。（）2个社会价值的来源不包括（）社会观念图书馆层面的障碍因素不去包括：（）书籍复杂无法分类下列不属于信息收集的过程的（）不提供信息收集的成果下列不属于信息收集的特点（）不定性

下列不属于用户在信息获取过程中的模糊性主要表现（）用户无法准确表达需求下列哪种方式不能做出信息商品（）口口相传下列现象不是五官感知信息为（）电波下列属于个人价值的来源（）个人需求信息information”一词来源于哪里（）拉丁信息管理的核心技术是（）信息的存储与处理技术信息管理观念是专业技术人员的（基本要素信息认知能力是专业技术人员的（基本素质）信息是通信的内容。控制论的创始人之一维纳认为，信息是人们在适应（外部）世界并且使之反作用于世界的过程中信息收集的对象是（信息源）由于（）是一种主体形式的经济成分，它们发展必将会导致社会运行的信息化。（信息经济）在网络环境下，用户信息需求的存在形式有（）4种判断题 “效益”一词在经济领域内出现，效益递增或效益递减也是经济学领域的常用词。（对） “知识经济”是一个“直接建立在知识和信息的生产、分配和使用之上的经济”，是相对于“以物质为基础的经济”而言的一种新型的富有生命力的经济形态。（对）百度知道模块是统计百度知道中标题包含该关键词的问题，然后由问

人教版高中语文必修一背诵篇目

高中语文必修一背诵篇目 1、《沁园春长沙》毛泽东独立寒秋，湘江北去，橘子洲头。看万山红遍，层林尽染；漫江碧透，百舸争流。鹰击长空，鱼翔浅底，万类霜天竞自由。怅寥廓，问苍茫大地，谁主沉浮？携来百侣曾游，忆往昔峥嵘岁月稠。恰同学少年，风华正茂；书生意气，挥斥方遒。指点江山，激扬文字，粪土当年万户侯。曾记否，到中流击水，浪遏飞舟？ 2、《诗两首》（1）、《雨巷》戴望舒撑着油纸伞，独自 /彷徨在悠长、悠长/又寂寥的雨巷，我希望逢着 /一个丁香一样的 /结着愁怨的姑娘。她是有 /丁香一样的颜色，/丁香一样的芬芳， /丁香一样的忧愁，在雨中哀怨， /哀怨又彷徨； /她彷徨在这寂寥的雨巷，撑着油纸伞 /像我一样， /像我一样地 /默默彳亍着冷漠、凄清，又惆怅。 /她静默地走近/走近，又投出太息一般的眼光，/她飘过 /像梦一般地， /像梦一般地凄婉迷茫。像梦中飘过 /一枝丁香的， /我身旁飘过这女郎；她静默地远了，远了，/到了颓圮的篱墙， /走尽这雨巷。在雨的哀曲里， /消了她的颜色， /散了她的芬芳， /消散了，甚至她的太息般的眼光， /丁香般的惆怅/撑着油纸伞，独自 /彷徨在悠长，悠长又寂寥的雨巷， /我希望飘过 /一个丁香一样的 /结着愁怨的姑娘。（2）、《再别康桥》徐志摩轻轻的我走了， /正如我轻轻的来； /我轻轻的招手， /作别西天的云彩。那河畔的金柳， /是夕阳中的新娘； /波光里的艳影， /在我的心头荡漾。软泥上的青荇， /油油的在水底招摇； /在康河的柔波里， /我甘心做一条水草！那榆阴下的一潭， /不是清泉，是天上虹 /揉碎在浮藻间， /沉淀着彩虹似的梦。寻梦？撑一支长篙， /向青草更青处漫溯， /满载一船星辉， /在星辉斑斓里放歌。但我不能放歌， /悄悄是别离的笙箫； /夏虫也为我沉默， / 沉默是今晚的康桥！悄悄的我走了， /正如我悄悄的来；/我挥一挥衣袖， /不带走一片云彩。 4、《荆轲刺秦王》太子及宾客知其事者，皆白衣冠以送之。至易水上，既祖，取道。高渐离击筑，荆轲和而歌，为变徵之声，士皆垂泪涕泣。又前而为歌曰：“风萧萧兮易水寒，壮士一去兮不复还！”复为慷慨羽声，士皆瞋目，发尽上指冠。于是荆轲遂就车而去，终已不顾。 5、《记念刘和珍君》鲁迅（1）、真的猛士，敢于直面惨淡的人生，敢于正视淋漓的鲜血。这是怎样的哀痛者和幸福者？然而造化又常常为庸人设计，以时间的流驶，来洗涤旧迹，仅使留下淡红的血色和微漠的悲哀。在这淡红的血色和微漠的悲哀中，又给人暂得偷生，维持着这似人非人的世界。我不知道这样的世界何时是一个尽头！

星巴克swot分析

6月21日 85度C VS. 星巴克SWOT分析星巴克SWOT分析优势 1.人才流失率低 2.品牌知名度高 3.熟客券的发行 4.产品多样化 5.直营贩售 6.结合周边产品 7.策略联盟劣势 1.店內座位不足 2.分店分布不均机会 1.生活水准提高 2.隐藏极大商机 3.第三空间的概念 4.建立电子商务威胁 1.WTO开放后，陆续有国际品牌进驻 2.传统面包复合式、连锁咖啡馆的经营星巴克五力分析１、供应商：休闲风气盛，厂商可将咖啡豆直接批给在家煮咖啡的消费者２、购买者：消费者意识高涨、资讯透明化（比价方便）３、同业內部竞争：产品严重抄袭、分店附近必有其他竞争者４、潜在竞争者：设立咖啡店连锁店无进入障碍、品质渐佳的铝箔包装咖啡５、替代品：中国茶点、台湾小吃、窜红甚快的日本东洋风...等８５度ｃ市場ｓｗｏｔ分析：【Strength优势】具合作同业优势产品精致以价格进行市场区分，平价超值服务、科技、产品、行销创新，机动性强加盟管理人性化【Weakness弱势】通路品质控制不易品牌偏好度不足通路不广财务能力不健全 85度c的历史资料，在他们的网页上的活动信息左邊的新聞訊息內皆有詳細資料，您可以直接上網站去查閱，皆詳述的非常清楚。

顧客滿意度形成品牌重於產品的行銷模式。你可以在上他們家網站找找看！【Opportunity机会】勇于變革变革与创新的经营理念同业策略联盟的发展、弹性空间大【Threat威胁】同业竞争对手(怡客、维多伦)门市面对面竞争加盟店水准不一，品牌形象建立不易直，间接竞争者不断崛起(壹咖啡、City Café．．．) 85度跟星巴克是不太相同的零售，星巴客应该比较接近丹堤的咖啡厅，策略形成的部份，可以从 1.产品线的宽度跟特色 2.市场区域与选择 3.垂直整合 4.规模经济 5.地区 6.竞争优势這6點來做星巴客跟85的区分可以化一个表，來比较他們有什么不同內外部的話，內部就从相同产业來分析(有什麼优势跟劣势) 外部的話，一樣是相同产业但卖的东西跟服务不太同，来与其中一个产业做比较(例如星巴客) S(优势):點心精致平价,咖啡便宜,店面设计观感很好...等等 W(劣势):对于消费能力较低的地区点心价格仍然较高,服务人员素质不齐,點心种类变化較少 O(机会):对于点心&咖啡市场仍然只有少数的品牌独占(如:星XX,壹XX...等),它可以透过连锁店的开幕达成高市占率 T(威协):1.台湾人的模仿功力一流,所以必须保持自己的特色做好市场定位 2.消費者的口味变化快速,所以可以借助学者"麥XX"的做法保有主要的點心款式外加上几样周期变化的點心五力分析客戶讲价能力（the bargaining power of customers）、供应商讲价能力（the bargaining power of suppliers）、新进入者的竞争（the threat of new entrants）、替代品的威协（the threat of substitute products）、现有厂商的竞争（The intensity of competitive rivalry）

2019年温州一般公需科目《信息管理与知识管理》模拟题

2019年温州一般公需科目《信息管理与知识管理》模拟题一、单项选择题(共30小题，每小题2分) 2.知识产权的时间起点就是从科研成果正式发表和公布的时间，但有期限，就是当事人去世（）周年以内权利是保全的。 A、三十 B、四十 C、五十 D、六十 3.（）国务院发布《中华人民共和国计算机信息系统安全保护条例》。 A、2005年 B、2000年 C、1997年 D、1994年 4.专利申请时，如果审察员认为人类现有的技术手段不可能实现申请者所描述的结果而不给予授权，这体现了（）。 A、技术性 B、实用性 C、创造性 D、新颖性 5.根据本讲，参加课题的研究策略不包括（）。 A、量力而行 B、逐步击破 C、整体推进 D、有限目标 6.本讲认为，传阅保密文件一律采取（）。 A、横向传递 B、纵向传递 C、登记方式 D、直传方式 7.本讲提到，高达（）的终端安全事件是由于配置不当造成。 A、15% B、35% C、65% D、95% 8.本讲认为知识产权制度本质特征是（）。 A、反对垄断 B、带动创业 C、激励竞争 D、鼓励创新 9.本讲指出，以下不是促进基本公共服务均等化的是（）。 A、互联网+教育 B、互联网+医疗 C、互联网+文化 D、互联网+工业 10.根据（）的不同，实验可以分为定性实验、定量实验、结构分析实验。 A、实验方式 B、实验在科研中所起作用C、实验结果性质D、实验场所 11.本讲认为，大数据的门槛包含两个方面，首先是数量，其次是（）。 A、实时性 B、复杂性 C、结构化 D、非结构化 12.2007年9月首届()在葡萄牙里斯本召开。大会由作为欧盟轮值主席国葡萄牙的科学技术及高等教育部主办，由欧洲科学基金会和美国研究诚信办公室共同组织。 A、世界科研诚信大会 B、世界学术诚信大会 C、世界科研技术大会 D、诺贝尔颁奖大会 13.（）是2001年7月15日发现的网络蠕虫病毒，感染非常厉害，能够将网络蠕虫、计算机病毒、木马程序合为一体，控制你的计算机权限，为所欲为。 A、求职信病毒 B、熊猫烧香病毒 C、红色代码病毒 D、逻辑炸弹 14.本讲提到，（）是创新的基础。 A、技术 B、资本 C、人才 D、知识 15.知识产权保护中需要多方协作，（）除外。 A、普通老百姓 B、国家 C、单位 D、科研人员 17.世界知识产权日是每年的（）。 A、4月23日 B、4月24日 C、4月25日 D、4月26日 18.2009年教育部在《关于严肃处理高等学校学术不端行为的通知》中采用列举的方式将学术不端行为细化为针对所有研究领域的()大类行为。 A、3 B、5 C、7 D、9 19.美国公民没有以下哪个证件（）。 A、护照 B、驾驶证 C、身份证 D、社会保障号 20.关于学术期刊下列说法正确的是（）。 A、学术期刊要求刊发的都是第一手资料B、学术期刊不要求原发 C、在选择期刊时没有固定的套式 D、对论文的专业性没有限制 22.知识产权的经济价值（），其遭遇的侵权程度（）。 A、越大，越低 B、越大，越高 C、越小，越高 D、越大，不变 23.（）是一项用来表述课题研究进展及结果的报告形式。 A、开题报告 B、文献综述 C、课题报告 D、序论 25.1998年，（）发布《电子出版物管理暂行规定》。

人教版高中语文必修必背课文

必修1 沁园春·长沙（全文）毛泽东独立寒秋，湘江北去，橘子洲头。看万山红遍，层林尽染，漫江碧透，百舸争流。鹰击长空，鱼翔浅底，万类霜天竞自由。怅寥廓，问苍茫大地，谁主沉浮。携来百侣曾游，忆往昔峥嵘岁月稠。恰同学少年，风华正茂，书生意气，挥斥方遒。指点江山，激扬文字，粪土当年万户侯。曾记否，到中流击水，浪遏飞舟。雨巷（全文）戴望舒撑着油纸伞，独自彷徨在悠长、悠长又寂寥的雨巷，我希望逢着一个丁香一样地结着愁怨的姑娘。她是有丁香一样的颜色，丁香一样的芬芳，丁香一样的忧愁，在雨中哀怨，哀怨又彷徨；

她彷徨在这寂寥的雨巷，撑着油纸伞像我一样，像我一样地默默彳亍着冷漠、凄清，又惆怅。她默默地走近，走近，又投出太息一般的眼光她飘过像梦一般地，像梦一般地凄婉迷茫。像梦中飘过一枝丁香地，我身旁飘过这个女郎；她默默地远了，远了，到了颓圮的篱墙，走尽这雨巷。在雨的哀曲里，消了她的颜色，散了她的芬芳，消散了，甚至她的太息般的眼光丁香般的惆怅。撑着油纸伞，独自彷徨在悠长、悠长又寂寥的雨巷，我希望飘过一个丁香一样地结着愁怨的姑娘。再别康桥（全文）徐志摩轻轻的我走了，正如我轻轻的来；我轻轻的招手，作别西天的云彩。那河畔的金柳，是夕阳中的新娘；波光里的艳影，在我的心头荡漾。软泥上的青荇，油油的在水底招摇；

在康河的柔波里，我甘心做一条水草！那榆荫下的一潭，不是清泉，是天上虹揉碎在浮藻间，沉淀着彩虹似的梦。寻梦？撑一支长篙，向青草更青处漫溯，满载一船星辉，在星辉斑斓里放歌。但我不能放歌，悄悄是别离的笙箫；夏虫也为我沉默，沉默是今晚的康桥。悄悄的我走了，正如我悄悄的来；我挥一挥衣袖，不带走一片云彩。记念刘和珍君（二、四节）鲁迅二真的猛士，敢于直面惨淡的人生，敢于正视淋漓的鲜血。这是怎样的哀痛者和幸福者？然而造化又常常为庸人设计，以时间的流驶，来洗涤旧迹，仅使留下淡红的血色和微漠的悲哀。在这淡红的血色和微漠的悲哀中，又给人暂得偷生，维持着这似人非人的世界。我不知道这样的世界何时是一个尽头！我们还在这样的世上活着；我也早觉得有写一点东西的必要了。离三月十八日也已有两星期，忘却的救主快要降临了罢，我正有写一点东西的必要了。四我在十八日早晨，才知道上午有群众向执政府请愿的事；下午便得到噩耗，说卫队居然开枪，死伤至数百人，而刘和珍君即在遇害者之列。但我对于这些传说，竟至于颇为怀疑。我向来是不惮以最坏的恶意，来推测中国人的，然而我还不料，也不信竟会下劣凶残到这地步。况且始终微笑着的和蔼的刘和珍君，更何至于无端在府门前喋血呢？然而即日证明是事实了，作证的便是她自己的尸骸。还有一具，是杨德群君的。而且又证明着这不但是杀害，简直是虐杀，因为身体上还有棍棒的伤痕。但段政府就有令，说她们是“暴徒”！但接着就有流言，说她们是受人利用的。惨象，已使我目不忍视了；流言，尤使我耳不忍闻。我还有什么话可说呢？我懂得衰亡民族之所以默无声息的缘由了。沉默啊，沉默啊！不在沉默中爆发，就在沉默中灭亡。

文档之家

COMBINING COMPETITIVE LEARNING NETWORKS OF VARIOUS REPRESENTATIONS FOR SEQUENTIAL DATA CLUS

最新The_Monster课文翻译

知识管理和信息管理之间的联系和区别_[全文]

新版人教版高中语文课本的目录。

信息管理与知识管理典型例题讲解

连锁咖啡厅顾客满意度涉入程度对忠诚度影响之研究以高雄市星巴克为例

高中外研社英语选修六Module5课文Frankenstein's Monster

人教版高中语文必修必背课文精编WORD版

(完整版)Unit7TheMonster课文翻译综合教程四

(完整版)信息管理与知识管理

人教版高中语文必修一背诵篇目

星巴克swot分析

2019年温州一般公需科目《信息管理与知识管理》模拟题

人教版高中语文必修必背课文