Social Behavioral Biometrics using Personality Traits-aware Tweet Embedding

Abstract

User recognition in online social networks has emerged as an important problem in the domain of social behavioral biometrics and social media forensics. In this thesis, the linguo-stylistic and semantic analysis of textual data used to predict a user's personality traits information is combined with the graph structure of social interaction to build a social behavioral biometric user recognition system. A Deep Neural Network (DNN) is first trained for the task of personality traits classification using only the textual content from online social network (OSN) profiles. Next, a novel weighted graph representation scheme is proposed to encode social interactions within OSNs, incorporating information regarding the psychological similarity between interacting users. Finally, a Graph Neural Network (GNN) is trained for the task of closed-set user recognition, and the utility of the proposed system is evaluated on two different datasets demonstrating its superiority to state-of-the-art methods aimed at user recognition.