Matching User Accounts across Social Networks based on Users Message

Identifying users across social networks has got more and more attention. The existing methods mainly estimate the pairwise similarity between users in different social networks and mainly rely on users' profiles and activities. But the users who pay attention to their privacy may change their profiles and relationships. In this paper, we propose a MUSIC (Modeling User Style for Identifying aCcounts across Social Networks) framework to address this problem: First, we build users content style model based on users message using word embedding technology; Second, we reduce the problem of finding users across social networks to classification problem on a single social network. Our experimental results validate the effectiveness and efficiency of our framework, and shows either all of user's message or only user's original posts can provide nearly the same efficiency in identifying this kind of users.