Sockpuppets Detection Using Markov-Driven Graph Convolutional Networks with Ensemble Learning

This paper introduces a novel framework for detecting sockpuppets in online social networks, utilizing an advanced approach that integrates Markov-driven Graph Convolutional Networks (MDGCN) with ensemble learning techniques. Our method capitalizes on the data inherent in social networks by applying GCNs to capture and learn from intricate user interactions. To address the class imbalance problem often encountered in sockpuppet detection, we incorporate Focal Loss into our training regimen, enhancing the model's capability to identify less frequent, harder-to-detect instances of false information spreaders. Additionally, we implement ensemble learning strategies to refine and bolster the predictive power of our model. Through extensive testing on multiple social network platform datasets, our proposed method demonstrates superior performance in identifying sockpuppets compared to existing techniques. This underscores its potential as a robust solution for safeguarding against the rapid spread of false information online.