| Recent years, heterogeneous information networks (HIN) has attracted much attentions by researchers from many domains, such as similarity search, classification, ranking and relationship prediction. Relationship prediction is a newly developed problem derived from link prediction which is a classical problem in homogeneous information networks. The most crucial part for relationship prediction in heterogeneous information networks (HIN) is how to effectively represent and utilize the information hidden in the creation of relationships.There exist three kinds of information that need to be considered, namely local structure information (Local-info), global structure information (Global-info) and attribute information (Attr-info). They influence relationship creation in a different but complementary way: Local-info is limited to the topologies around certain nodes thus it ignores the global position of node; methods using Global-info are biased to highly visible objects; and Attr-info can capture features related to objects and relations in networks. Therefore, it is essential to combine all the three kinds of information together. However, existing approaches utilize them separately or in a partially combined way since effectively encoding all the information together is not an easy task. In this paper, a novel three-phase Supervised Ranking framework (S-Rank) is proposed to tackle this issue.To the best of our knowledge, our work is the first to completely combine Global-info, Local-info and Attr-info together. Firstly, a Supervised PageRank strategy (SPR) is proposed to capture Global-info and Attr-info. Secondly, we propose a Meta Path-based Ranking method (MPR) to obtain Local-info in HIN. Finally, they are integrated into the final ranking result. Experiments on DBLP data demonstrate that the proposed S-Rank framework can effectively take advantage of all the three kinds of information for predicting citation relation and outperforms other well-known baseline approaches. |
PDF Full Text Request