COMPOSITE BEHAVIOURAL MODEL FOR IDENTIFY THEFT DETECTION IN ONLINE SOCIAL NETWORK

Abstract

This study aims to develop an efficient and responsive behavioral model for detecting 
online identity theft, particularly focusing on online social networks (OSNs) where 
users' behaviors are multifaceted and encompass various low-quality data types like 
offline check-ins and online user-generated content (UGC). Through our investigation, 
we confirm the synergistic effect of integrating different dimensions of user records to 
model their behavioral tendencies effectively. To leverage this synergy, we propose a 
novel joint modeling approach that captures both online and offline features of users' 
composite behavior. Evaluation of our joint model against traditional models and their 
fused counterparts on real-world datasets from Foursquare and Yelp demonstrates 
superior performance, with AUC values of 0.956 and 0.947, respectively. Notably, 
our model achieves a recall rate of 65.3% in Foursquare and 72.2% in Yelp, with a 
minimal false-positive rate below 1%. Importantly, these results are obtained with 
minimal response latency, as our method requires the examination of only one 
composite behavior. This research sheds light on enhancing real-time online identity 
authentication through a deeper understanding of users' composite behavioral patterns, 
offering valuable insights to the cybersecurity community.