Keyword Associaiton Linked Network-based Web Event Evolution Analysis

With the rapid development of social informatization and the influences from domestic and overseas, the public security events have become an increasingly important problem in China, which makes it a great challenge to the relevant departments. Furthermore, the adventure and extensive usage of the Web, on the one hand, accelerates of the evolution of the public security events and enhances their impact. On the other hand, Web also provides an efficient tool for the public security event analysis. Therefore, this study focuses on the research of the public security events based on the large-scale webpages on the web.At first, we build a Keyword Association Linked Network(KALN) as a complete and unified representation of an event for capturing the semantics scattered in large-scale webpages. The KALN lays the foundation for the further events analysis in this study.On the one hand, based on the KALN, we study the hidden topics in an event and the preferences of different websites on these topics. More specially, 1) a new graph-based representation method for the webpages is proposed based on the KALN, which can significantly improve the capability of the semantic capture. In order to adapt to the new representation method, a new topic model is proposed to discovery the hidden topics in a web event. The experiments show that the new model outperforms the traditional one based on the VSM. 2) website preferences on the hidden topics are investigated based on the website-webpage-keyword network. Two strategies are proposed: one is to explicitly utilize the hierarchical network by the community detection; the other is to implicitly utilize the hierarchical network by the topic model.On the other hand, based on the KALN, we also study the evolution of the web event. More specially, 1) the uncertainty of the web event is defined and measured by the entropy of keyword weight distribution where three strategies are proposed to define keyword weights: the first is based on statistical property; the second is the combination of statistical property and local structural property; the third is the combination of statistical property and global structural property. A semantic pyramid is then constructed considering the hierarchy of the keyword uncertainty, which is applied to the webpage recommendation. 2) An uncertainty space is built based on the entropy and power-law distribution of the KALN with two extreme states: the most uncertain state and the most certain state. According to the position of the state of current web event in this uncertainty space, we define and evaluate the inner evolution power of the web event. 3) A social context model is built referencing the Social Schema Theory and Anchor Theory for the measurement of the influence to the evolution of web event from the social environment. We define this kind of influence as the outer evolution power of the web event. Experiments on the real-world dataset prove the effectiveness of defined powers.Finally, this study can be used not only to provide new web services for helping users understand web events and their evolution, but also to help the relevant departments timely and accurately make policies to response to the web events in order to save the money and reduce the costs due to the events.