Study On Trust Model And Controversy Discovery Under Web2.0 Circumstance

The models for information generation, storage, dissemination and sharing are varied because of the wide applications of Web2.0 technologies, for examples P2P, blog, social network and wiki. Meanwhile, Web2.0 provides convenience for Web users’work and life. However, users may distrust Web content or the communication party because Internet is open, virtual and anonymous. Thus they refuse to use some web applications. Furthermore, the disputes among different users are produced for participation and diffusion properties of Web2.0, which affects the trust degree of Web content and information dissemination. So trust evaluation, trust management and controversial content detection under web2.0 circumstance become the research hotspots in computer field.In this thesis, the existing trust models and controversy discovery algorithms are compared and analyzed based on the introduction of main technologies and features of Web2.0, trust concepts, properties, measure methods and management model. Then according to the questions about differences among users’subjective evaluations and inconformity between text comments and ratings, several solutions and algorithms are proposed, which includes trust model based on HMM and sentiment analysis, trust evaluation method based on AHP for e-commerce, e-commerce reputation model based on elimination differences of user subjective evaluation. Finally a controversial degree algorithm based on text differences is designed to rank controversial issues of a wiki system, and the background color of controversial words is changed to visualize an article’s controversy. The specific studies are as follow:(1) An HMM and sentiment analysis based trust model is constructed. In terms of trust subjectivity, the discrete HMM model is used to describe the dynamic trust of an entity. And trust degrees are treated as hidden states. The text comments are transformed to trust ratings through sentiment polarity classification. A synthetical evaluation generating algorithm is proposed by combining original ratings. Trust value of an entity is computed based on HMM and synthetical evaluations as observation sequence. Finally, satisfactory results are obtained with simulated experiments. The precision of trust computation is improved, and trust dynamics is reflected(2) An efficient approach for evaluating trust in e-commerce is designed. Based on analyzing the factors which affects the trust of a buyer to a seller, a hierarchical evaluation index system for website quality is established, and the weight value of each index is computed by using the AHP method. The computation methods for initial trust and repeat trust are given according to electronic trading features and trust influencing factors. Lastly, a case is studied by using these evaluation methods, which verifies the practicality and effectiveness of our methods.(3) E-commerce reputation model based on elimination differences of user subjective evaluation is created. In order to enhance the objectivity and accuracy of reputation value in e-commerce, the algorithm for eliminating differences of users’subjective evaluation is designed by using trust translation. Then a dynamic e-commerce reputation management model is also constructed. Finally, the results of our experiments demonstrate the accuracy and validity of our method.(4) Controversial content discovery algorithm for Wikipedia based on text differences is proposed. Firstly, a malicious modification filtering algorithm is designed by using paragraphs, words, and references statistics of Wikipedia article’s all revisions. Then a statistic algorithm based on peak-valley detection is developed to confirm which revision is the mature sign for a wiki article. The controversial words are obtained through counting different strings between any two consecutive revisions. And the algorithms for computing article controversy and user controversy are given. Meanwhile background color of the most debatable words is varied to highlight the controversial information. The dataset collected from English Wikipedia are used to evaluate the effectiveness of the proposed method. The experimental results show a very promising performance advantage in detecting controversial information. In addition, a wiki site is created with Mediawiki engine to display results.