Hypernetwork-based Tags Similarity Measur--ement For Social Tagging Systems

In recent years, the emergence of social tagging effectively improved the efficiency of resources retrievaling and using. And it has gradually developed into an important resources sharing platform. Measuring tags similarity is the foundation of link prediction and personalized recommendation for social tagging systems. In order to solve the problem in existing tags similarity methods based on space vector matrix, bipartite graph, tripartite graph, tag co-occurrence network and so on, which split the internal relations of “users-items-tags” in social tagging system. Combined with the Natural Science Fund Project “Research on social-driven context-aware personalized information service in ubiquitous computing environment”(project NO.:71471165) and the Social Science Planning Project of Ministry of Education of China “Personalized information service research in mobile environment—a context-aware and ontology based approach”(project NO.: 14JYC870010), an approach to measuring social tags similarity based on hypernetwork is put forward. After that, the method of constructing the hypernetwork for social tags is elaborated systematically and a series of measures for social tags similarity is proposed. Then the main work of this thesis is as follows:(1) Based on the hypergraph theory, this paper constructs the hypernetwork of social tags. The hypernetwork abstract users’ tagging behaviors as hyper edges and the tags it used as nodes. It can not only accurately depict the structure of “user-item-tag” in social tagging system, but also avoid the unnecessary information losing in the construction of model effectively.(2) Combining with tagging relationship between tags in the reality system, a series of measures for social tags similarity is proposed, which follows principles of the number of common hyper-edges and the number of nodes in one hyper-edge based on hypernetwork, and then computer implementation is given.(3) We evaluate our approach on data sets from two real-world social tagging systems: Delicious and Last.fm.By incorporating the AUC and Precision criterions in link prediction, the experimental results show that the proposed tags similarity measures following defined principles perform well and are much better than the similarity measures based on tag co-occurrence network in prediction accuracy. This research result has great practical application value for link prediction and personalized recommendation based on social tag.