Enhanced Rumor Detection Model for Weibo based on Dual-Layer Multi-Head Attention Mechanism

Abstract

The spread of rumors on Weibo severely threatens the security of the cyberspace information ecosystem. Their deceptive nature complicates the dynamic coupling mechanism of multi-dimensional semantic features in rumor detection. To address this issue, this paper proposes an enhanced rumor detection model based on a dual-layer multi-head attention mechanism. The model employs a hierarchical strategy to extract key semantic features from word-level to sentence-level in Weibo texts. Additionally, during word-level encoding, it leverages a rumor lexicon to further enhance the semantic features of relevant words, mitigating the feature sparsity problem in rumor detection to some extent. Experimental results demonstrate the model's effectiveness in optimizing feature selection and improving rumor detection performance.