LINKSOCIAL: Linking User Profiles Across Multiple Social Media Platforms

Abstract

Social media connects individuals to on-line communities through a variety of platforms, which are partially funded by commercial marketing and product advertisements. A recent study reported that 92% of businesses rated social media marketing as very important. Accurately linking the identity of users across various social media platforms has several applications viz. marketing strategy, friend suggestions, multi platform user behavior, information verification etc. We propose LINKSOCIAL, a large-scale, scalable, and efficient system to link social media profiles. Unlike most previous research that focuses mostly on pair-wise linking (e.g., Facebook profiles paired to Twitter profiles), we focus on linking across multiple social media platforms. L INK S OCIAL has three steps: (1) extract features from user profiles and build a cost function, (2) use Stochastic Gradient Descent to calculate feature weights, and (3) perform pair-wise and multi-platform linking of user profiles. To reduce the cost of computation, L INK S OCIAL uses clustering to perform candidate pair selection. Our experiments show that L INK S OCIAL predicts with 92% accuracy on pair-wise and 74% on multi-platform linking of three well-known social media platforms. Data used in our approach will be available at http://vishalshar.github.io/data/.