Enhanced Multi-Relational Graph Representation for Incremental Social Event Detection

Social event detection have been recognized as an important tool for obtaining information. Its main function is to recognize significant events in social media information. Current graph neural network approaches are limited by three key challenges: text sparsity, feature ambiguity, and data stream timeliness. To address these limitations, we introduce an innovative solution called Multi-Relational Graph Represen-tation Enhanced Incremental Social Event Detection Model (MRGEvent). Multidimensional feature fusion is achieved by combining homogeneous and weighted two-channel graph contrastive learning with an attention mechanism. Adaptive incremental clustering is designed using reinforcement learning principles to accommodate data stream dynamics. The final experimental results show that the proposed model exhibits significant performance improvement in complex social scenarios.