Accurate online social network user profiling

Abstract

We present PGPI+ (Partial Graph Profile Inference+) an improved algorithm for user profiling in online social networks. PGPI+ infers user profiles under the constraint of a partial social graph using rich information about users (e.g. group memberships, views and likes) and handles nominal and numeric attributes. Experimental results with 20,000 user profiles from the Pokec social network shows that PGPI+ predicts user profiles with considerably more accuracy and by accessing a smaller part of the social graph than five state-of-the-art algorithms.