Social Identity Link Across Incomplete Social Information Sources Using Anchor Link Expansion

Abstract

Social link identification SIL, that is to identify accounts across different online social networks that belong to the same user, is an important task in social network applications. Most existing methods to solve this problem directly applied machine-learning classifiers on features extracted from user’s rich information. In practice, however, only some limited user information can be obtained because of privacy concerns. In addition, we observe the existing methods cannot handle huge amount of potential account pairs from different OSNs. In this paper, we propose an effective SIL method to address the above two challenges by expanding known anchor links (seed account pairs belonging to the same person). In particular, we leverage potentially useful information possessed by the existing anchor link, and then develop a local expansion model to identify new social links, which are taken as a generated anchor link to be used for iteratively identifying additional new social link. We evaluate our method on two most popular Chinese social networks. Experimental results show our proposed method achieves much better performance in terms of both the number of correct account pairs and efficiency.