This paper presents the coauthor network topic (CNT) model constructed based on Markov random fields (MRFs) with higher-order cliques. Regularized by the complex coauthor network structures, the CNT can simultaneously learn topic distributions as well as expertise of authors from large document collections. Besides modeling the pairwise relations, we model also higher-order coauthor relations and investigate their effects on topic and expertise modeling. We derive efficient inference and learning algorithms from the Gibbs sampling procedure. To confirm the effectiveness, we apply the CNT to the expert finding problem on a DBLP corpus of titles from six different computer science conferences. Experiments show that the higher-order relations among coauthors can improve the topic and expertise modeling performance over the case with pairwise relations, and thus can find more relevant experts given a query topic or document.

Opinion mining is of great significance in the analysis of user generated content. While there is some progress in supervised classification of opinion, the unsupervised learning of product features has drawn less attention. Unlike previous approaches based on basic syntactic pattern, our product feature mining utilizes syntactic dependency knowledge in a novel way by discriminating nominal and non-nominal terms. A nominal semantic structure will be parsed based on a dependency tree together with our model treating non-nominal terms as the semantic neighbors of the associated nominal terms. The semantic structure parsing will produce an opinionated pair stream with couples of nominal terms and its semantic neighbors, based on which fine-grained product features can be obtained by co-clustering approach via factorization method. Evaluation on average cluster entropies, perplexity and manual evaluation demonstrated advantage of our model. Product features highly cohesive in fine-grain are extracted automatically.

In this paper we propose the multirelational topic model (MRTM) for multiple types of link modeling such as citation and coauthor links in document networks. In the citation network, the MRTM models the citation link between each pair of documents as a binary variable conditioned on their topic distributions. In the coauthor network, the MRTM models the coauthor link between each pair of authors as a binary variable conditioned on their expertise distributions. The topic discovery is collectively regularized by multiple relations in both citation and coauthor networks. This model can summarize topics from the document network, predict citation links between documents and coauthor links between authors. Efficient inference and learning algorithms are derived based on Gibbs sampling. Experiments demonstrate that the MRTM significantly outperforms other state-of-the-art single-relational link modeling methods for large scientific document networks.

Real-world social networks (e.g., blogosphere) often evolve over time and thus poses challenges on conventional social network analysis techniques which model the underlying networks as static graphs. In this paper, we are interested in detecting dynamic communities and their trend of evolution in a social network by examining the structural and dynamic patterns of interactions. In doing so, we propose an iterative mining algorithm for computing the intensities and bursts of some hidden communities over time. Our method is probabilistic in nature and can be applied to both undirected graphs and directed graphs. Quantitative and qualitative performance comparisons between the proposed method and some representative methods for social network analysis are provided. Evaluation results based on three benchmark datasets, including Reuters terror news network, political blogosphere, and Enron emails, show that the proposed method is both effective and efficient.

In Web forum analysis, both the discussion topics and author interests are greatly concerned. We introduce a linked topic and interest model based on latent Dirichlet allocation (LDA) to explore discussion topics and author interests. Rather than having two separate models or modeling combined topics and interests with just one hidden topic assignment variable, the proposed model has separate but linked hidden variables for topic and interest exploration. As exact model parameter inference is intractable, Gibbs sampling is employed to estimate topic, author, and interest distributions. The joint distribution of the linked hidden variables also provides an interpretation of an interest in terms of weighted topics or vice versa. We apply the model to a NIPS data set and a corpus containing text contents of a popular digital camera Web forum. Topics and interests discovered by using the model is demonstrated. The model generalization capability is also assessed by means of perplexity and the results show that the linked topic and interest model has performance exceeding that of LDA document topic model and author topic model.

Social media such as Web forum often have dense interactions between user and content where network models are often appropriate for analysis. Joint non-negative matrix factorization model of participation and content data can be viewed as a bipartite graph model between users and media and is proposed for analysis social media. The factorizations allow simultaneous automatic discovery of leaders and sub-communities in the Web forum as well as the core latent topics in the forum. Results on topic detection of Web forums and cluster analysis show that social features are highly effective for forum analysis.

Location estimation using the Global System for Mobile communication (GSM) is an emerging application that infers the location of the mobile receiver from multiple signals measurements. While geometrical and signal propagation models have been deployed to tackle this estimation problem, the terrain factors and power fluctuations have confined the accuracy of such estimation. Using support vector regression, we investigate the missing value location estimation problem by providing theoretical and empirical analysis on existing and novel kernels. A novel synthetic experiment is designed to compare the performances of different location estimation approaches. The proposed support vector regression approach shows promising performances, especially in terrains with local variations in environmental factors.

The proliferation of unsolicited emails, also known as spam, poses significant burden to email users worldwide. Recent researches on spam filtering have shown that high accuracies can be obtained if labeled emails examples are available from the particular user of the spam filter. However, the time consuming process of providing personalized labeled training examples is often inconvenient or impossible due to privacy issues. In this paper, a semi-supervised personalized spam filter based on classifier ensemble is proposed that classifies user's emails accurately by learning on both generic labeled emails and personalized unlabeled emails. The proposed multi-stage classification process begins learning a SVM model from labeled generic data. Unlabeled user's emails are then fed to this SVM to generate personalized labeled data for constructing personalized naive Bayes classifiers. Furthermore, some personalized labeled examples are generated by exploiting rare word distributions and then fed into a semi-supervised classifier. The multi-stage results are integrated with SVMs learned from generic labeled emails to produce the final classification results. Experimental results show that the proposed approaches can significantly increases the classification accuracy in spam filtering

The interaction between usability and how a Web site is structured is a complicated issue. In this paper, we discuss a Web structure mining algorithm which allows the automatic extraction of navigational structures in a Web site without performing hypertext analysis. We perform several usability experiments to correlate the usability of Web sites and the structural design of the Web site. Experimental results show that the structure mining algorithm gives reasonable prediction about several design issues in Web structure. The analysis serves as building block in the complex issue of web usability and structure mining.

Recently, mobile location estimation is drawing considerable attention in the field of wireless communications. Among different mobile location estimation methods, the one which estimates the location of mobile stations with reference to the wave propagation model is drawing much attention. This approach, in principle, makes use of the most primitive property of wave propagation - signal strength, to perform location estimation. Hence this approach should be able to apply to different kinds of cellular network. We found out that in estimating mobile location with reference to signal strength, the azimuth gain of directional antenna and environmental factors can help to improve the accuracy. In this paper, we report our study of a directional propagation model (DPM) which enhanced the traditional propagation model with these factors. We experimented our model with 3,703 sets of real life data collected from a major mobile phone operator of Hong Kong. Results show that the DPM models have significant improvement over other existing location methods in terms of accuracy and stability.

The paper describes a novel watermarking method to hide a binary watermark in image files compressed by fractal block coding. The watermarking method utilizes a special type of orthogonalization fractal coding method where the fractal affine transform is determined by range block mean and contrast scaling. Such orthogonalization fractal decoding has a unique mean-invariant property for each range block where the mean of a range block of the decoding series is invariant. The proposed watermark embedding procedure inserts an m-sequence into the quantized code of the range block means. The watermark detection is statistically determined by computing the correlation coefficient. Experimental results show that the watermark is robust against image processing operations such as JPEG compression, low-pass filtering, rescaling and clipping.

A texture image can be characterized by its fractal feature, and hence, the fractal features can be used as the texture signature to retrieve the images. In this paper, based on the independence of fractal parameters, we improve the three-dimensional histogram into a composite index, in order to reduce computational complexity. Furthermore, under the same computational cost, we make the comparison between the proposed composite index and fractal-dimension-related features, and experimental results on a database of 640 texture images indicate that the proposed composite index outperforms fractal-dimension-related features

This paper describes the issues involved in designing and implementing a large-scale cooperative object database server for collaborative virtual environment. Participants presented in the environment may include humans and computer artifacts. The focus of this paper is the distributing factor and communication model between nodes. In DIVINE, all objects will be decentralized and scattered among three tiers of object databases. By limiting knowledge and refining tasks for each server the network and processor workload can be reduced. The objective of the system is to provide a virtual environment for distributed computing with N+1 fault tolerant and subject to expansion without interruption of services.

A regularized color clustering algorithm is proposed to solve the color clustering problem in medical image database. By incorporating both measures of cluster separability and cluster compactness, regularized color clustering allows the automatic extraction of significant color groups with varying populations. Experimental results in different color spaces show that the regularized color clustering gives superior results in extracting significant distinct/abnormal color clusters without significant increases in cluster compactness. Furthermore, results of color clustering in different color spaces show that the LUV color space is more suitable for color clustering. Methods for selecting the regularization constants have also been suggested.