科研成果详情

The rapid propagation of rumors on social media can give rise to various social issues, underscoring the necessity of swift and automated rumor detection. Existing studies typically identify rumors based on their textual or static propagation structural information, without considering the dynamic changes in the structure of rumor propagation over time. In this paper, we propose the Temporal Tree Transformer model, which simultaneously considers text, propagation structure, and temporal changes. By analyzing observing the growth of propagation tree structures in different time windows, we use Gated Recurrent Unit (GRU) to encode these trees to obtain better representations for the classification task. We evaluate our model’s performance using the PHEME dataset. In most existing studies, information leakage occurs when conversation threads from all events are randomly divided into training and test sets. We perform Leave-One-Event-Out (LOEO) cross-validation, which better reflects real-world scenarios. The experimental results show that our model achieves state-of-the-art accuracy 75.84% and Macro F1 score of 71.98%, respectively. These results demonstrate that extracting temporal features from propagation structures leads to improved model generalization.