Balancing the Scales: HyperSMOTE for Enhanced Hypergraph Classification

With the proliferation of bots on social media platforms, especially X, the need for effective and efficient bot detection mechanisms has never been more paramount. However, the inherent imbalance between the number of genuine users and bots presents a significant challenge, often leading to biased classifiers. In this paper, we introduce HyperSMOTE, a novel approach for imbalanced node classification leveraging the rich structure of hypergraphs. By representing X users as nodes and their interactions as hyperedges, we construct a hypergraph that captures the intricate relationships and interactions among users. This hypergraph-based representation allows for a more nuanced understanding of user behavior and interactions, providing a robust foundation for bot detection. HyperSMOTE addresses the class imbalance by generating synthetic bot accounts in the hypergraph, ensuring a balanced training dataset while preserving the hypergraph’s semantics. Our method significantly outperforms existing baselines across various metrics, demonstrating its efficacy. We further delve into the impact of different upsampling scales on classification performance, providing insights into the optimal configurations for HyperSMOTE.