BASS: A Bootstrapping Approach for Aligning Heterogenous Social Networks

摘要

Most people now participate in more than one online social network (OSN). However, the alignment indicating which accounts belong to same natural person is not revealed. Aligning these isolated networks can provide united environment for users and help to improve online personalization services. In this paper, we propose a bootstrapping approach BASS to recover the alignment. It is an unsupervised general-purposed approach with minimum limitation on target networks and users, and is scalable for real OSNs. Specifically, we jointly model user consistencies of usernames, social ties, and user generated contents, and then employ EM algorithm for the parameter learning. For analysis and evaluation, We collect and publish large-scale data sets covering various types of OSNs and multi-lingual scenarios. We conduct extensive experiments to demonstrate the performance of BASS, concluding that our approach significantly outperform state-of-the-art approaches.