Heterogeneous Social Network Graph Matching using Structural Features

Abstract

Social Information from social networks has been used in diverse research areas. Since social networks can provide abundant information, employment of social information commonly regards as the solution of data sparsity problem. In recommender system, for example, numerous researchers uses social information to solve cold start problem, which is that the system cannot draw any inferences for object who has not yet gathered sufficient information. However the information provided by one social network is very limited to surmount data sparsity problem. Graph matching techniques which combines information of heterogeneous social network can be broad and firm base of ii other social network related research areas. Nowadays, users are opt to involve in multiple social networks simultaneously. Due to the fact that each social networks offer distinct service function and that data published for research is usually anonymized, there are not sufficient common information among heterogeneous social networks services. However, the graph structure formed by a same user tends to remain similar. In light of above, we propose novel approach to integrate heterogeneous social networks. Differ from other heterogeneous graph matching, we use not only simple in-and-out degree of social networks, but also Jaccard coefficient, Adamic/Adar score, Clustering coefficient, and Page rank to evaluate social status of user. Extensive experiments conducted on multiple real-world data and prove that our proposed method outperforms existed graph matching algorithm.