RISK ASSESSMENT IN SOCIAL NETWORKS BASED ON USER ANOMALOUS BEHAVIORS

Abstract

Online Social Networks (OSNs) have become integral to modern communication, enabling users to connect, share information, and engage globally. However, the open nature of these platforms exposes them to various security and privacy risks from users exhibiting anomalous behaviors, such as fake accounts, compromised profiles, spam campaigns, coordinated misinformation, and cyberbullying. Risk Assessment in Social Networks Based on User Anomalous Behaviors is a machine learning-driven framework designed to detect and quantify risks associated with deviant user activities. The system analyzes user behavioral patterns (posting frequency, interaction graphs, content semantics, and temporal activity) to identify deviations from normal behavior. It employs unsupervised anomaly detection techniques, including clustering of similar users, behavioral modeling, and metrics like deviation scores to assign risk levels to individual users or groups. By grouping users with similar profiles (based on demographics, activity levels, and network structure) and building group-specific normal behavior models, the framework achieves higher accuracy than global models. Experiments on real-world datasets (e.g., Facebook-like graphs) demonstrate effective detection of diverse anomalous behaviors such as Sybil attacks, click-spam, and collusion networks while maintaining low false positives. This approach enhances platform security, reduces malicious activity, and supports proactive risk mitigation for OSN providers and users.