TRANSFORMERS-BASED FAKE SOCIAL MEDIA PROFILE DETECTION USING DEEP LEARNING

Abstract

The rapid growth of social media platforms has made them a fertile ground for the spread of fake accounts, misinformation, and coordinated bot activity. Traditional detection methods often struggle to capture complex contextual and sequential patterns in user behaviour and content. This paper proposes a Transformer-based framework for detecting fake users and malicious content on social networks. By leveraging self-attention mechanisms, the model analyses both the temporal sequence of user activities and the semantic relationships in posts, comments, and messages, identifying subtle anomalies and coordinated behaviours. The system integrates textual, behavioural, and network-level features to enhance detection accuracy. Experimental results demonstrate that Transformer-based models outperform classical machine learning and graph-based methods, achieving higher F1-scores in detecting fake accounts, botnets, and misleading content.