Learning a Probabilistic Semantic Model from Heterogeneous Social Networks for Relationship Identification

Nowadays, social networks play an important role in our lives, in which information and knowledge are exchanged, shared and transformed. With time, large volumes of real-world data have been accumulated capturing diversified application domains. However, heterogeneity and incompleteness of data make social networks perform as ´data isolated islands´ separated to each other. In this paper, we propose a generic approach that consists of an ontology-based social network integration approach and a statistic learning method towards the Semantic Web data. In particular, an extended FOAF (Friend-Of-A-Friend) ontology is used as the mediation schema to integrate social networks and a hybrid entity reconciliation method is used to resolve entities of different data sources. We also present an analyzing approach that learns a probabilistic semantic model (PSM) from social data for relationship identification. Empirical results prove that, compared with single numeric or logical methods respectively, our hybrid reconciliation method has obvious improvements on precision and recall during combining LinkedIn.com and DBLP. In addition, PSM framework can reserve richer semantics of semantic data completely during data analysis.