A Study On Web Social Network-based Collaborative Filtering Model

As increasing development and pervasiveness of Internet, people are trapped and puzzled by large amount of information. It becomes more and more difficult for them to obtain just needed information from the fast-growing and large-scale information. Particularly, since end of 1990s, a large number of sites, namely Blog, based on automatic Web information publishing system appeared. These sites updated frequently and published with various personalized information. In near future, it's no doubt that they'll become another major source of information overload.Currently, people mainly depend on information retrieval technology, such as keyword-based search engines, to access useful or personal-like Blog information. However, there are at least three drawbacks or limitations of this approach. Firstly, people have to state their information needs explicitly;secondly, information published on Blog are usually subjective, which makes it hard to take quality of information into account;thirdly, varieties of Blog information makes it hard to represent in keywords. Collaborative filtering-based recommender system, which had been applied successfully in E-Commerce domain, could filter various types of information and recommend information beyond people's ever interests. But while being referred to recommend Blog information, collaborative filtering can not be easily planted, for Blog information usually scattered on different Web servers whichis difficult to be managed centrally.Regarding problem stated above, collaborative filtering-based recommending model, whose key is neighborhood formation, was thoroughly studied in this dissertation. Based on strong interactions among Blogs, applying semantic Web technology, a Blog-oriented, Web social network based collaborative filtering model was proposed. This model has six components in all. They are separately are RSS aggregator, user modeling, Web social network, RDF storing and processing, and RDF collecting. Functions and relating algorithms of these components were also given. Finally, using real data collected from the Web, experiments were designed and implemented. Empirical results proved the feasibility and efficiency of the proposed model.