Research On Tag Similarity In Social Tagging

The ternary relationship among users?resources and tags in social tagging system is formed by tagging resources with tags.As a link between users and resources,tags can be used not only to describe users' interests and hobbies,but also represent the content characteristics of resources.Tag similarity measurement has become the basis and core of tag applications in such as tag-based personalized recommendation and tag clustering.Therefore,how should be designed the effective tag similarity indicators by using the ternary relationship among users,resources and tags has become a major problem.For this reason,this thesis combines with the project of National Natural Science Foundation of China: "Research on Social-Driven Context-Aware Personalized Information Service in Ubiquitous Computing Environment"(Project No.71471165),from the perspective of user annotation behavior,a tag hypernetwork is constructed to depict the ternary relationship among users,resources and tags.On this basis,the influence of co-occurrence relationship with superedge and superedge co-occurrence relationship with superedge co-nodes on tag similarity in tag hypernetwork is considered comprehensively.A tag similarity evaluation method based on co-occurrence relationship in hypernetwork is proposed to measure tag similarity effectively.The main work and contributions of this thesis are as follows:(1)The tag hypernetwork is constructed.Focus of on user taging behavior,a tag hypernetwork which can clearly depict the ternary relationship among users?resources and tags in social tagging system is constructed and compared with tag co-occurrence network.(2)A set of tag similarity index based on co-occurrence relationship in hypernetwork is proposed.In this thesis,the three-dimensional closure principle in social network analysis and the adjacent association rule in psychology are incorprated tag hypernetwork.A set of tag similarity indexes based on the co-occurrence relationship of tag hypernetwork are constructed.considering the influence of co-occurrence relationship with super-edge and super-edge co-occurrence relationship on tag similarity.(3)Empirical study is carried out from data sets in last.fm and Movie Lens to test similarity index by using tag spectrum clustering.The contour coefficient of internal evaluation index in unsupervised clustering is utilized to verify the effectiveness of tag similarity indexes based on cooccurrence relationship in hypernetwork.The results show that the tag similarity index based on co-occurrence relationship in hypernetwork is effective,and more accurate than that based on co-occurrence relationship with super-edge in hypernetwork in the judgment of tag similarity relation.The research results can be applied better for personalized recommendation and clustering based on tags.