Evaluating Quality Of Web2.0Ugc Based On User Interaction Relationship

With the rapid development of Web2.0, there has been more and more websites whose information resource depends on user’s involvement and construction. These information resources are called user-generated content (UGC) which created by users and constantly enrich and expand in the communication process. Currently, researches dependent on UGC have been supported by the industry-wide concern and have made a lot of valuable research results. UGC could be composed and published by every user, which is a distinct character of Web2.0compared with Web1.0. As a result, quality of UGC may not be well-guaranteed. When being used in many applications, such as Information Retrieval, Data Mining, Topic Detection and Tracking, UGC itself demands a reasonable quality evaluation.In this thesis, we analyze the interaction relationship between UGC users, measure the extent of the authority of the user, and then put forward a synthetic evaluation method for UGC quality of content via user authority and topic distribution. As to user authority, we consider various features of users, including the basic registration information, the relative static online social relations and dynamic interactions between users. Then we build user relation networks and analyze the social nature of the network nodes in the process of social network analysis. We adopt Link Analysis based on user credibility and neural network regression analysis to rate user authority. As to topic distribution, there are multiple authors contributing to the generation of UGC. We analyze the probabilistic topic distribution of authors and UGC based on textual content using Author-Topic Model (AT). We’ve also integrated the author’s actual behavior during the composing of UGC to evaluate the overall quality of a multi-author UGC.We collect a large-scale real data from the TianYa Forum and conduct several comparative experiments based on various factors. A more convincing assessment method show that the proposed method could evaluate UGC reasonably.This work was supported in part by the National Science Foundation of China (NSFC) under Grants71231002and61202247.