Early Detection of Malicious Accounts on Social Platforms Based on Temporal Graph Feature Learning

Abstract

Social media platforms face escalating threats from malicious accounts that engage in disinformation campaigns, fraudulent activities, and coordinated attacks. Traditional detection methods often fail to identify these accounts before significant damage occurs. This paper presents a novel approach for early detection of malicious accounts using temporal graph feature learning. Our methodology constructs dynamic social graphs incorporating temporal behavioral patterns and employs graph neural networks to extract multi-dimensional features characterizing account evolution over time. The proposed framework analyzes registration patterns, social interaction dynamics, and content publishing behaviors to identify suspicious accounts in their early stages. Experimental results demonstrate superior performance compared to existing baseline methods, achieving 94.7% accuracy and 91.3% F1-score while maintaining real-time processing capabilities for large-scale social networks.