From Detection to Explanation: Integrating Temporal and Spatial Features for Rumor Detection and Explaining Results Using LLMs

The proliferation of rumors on social media has caused serious harm to society. Although previous research has attempted to use deep learning methods for rumor detection, they did not simultaneously consider the two key features of temporal and spatial domains. More importantly, these methods struggle to automatically generate convincing explanations for the detection results, which is crucial for preventing the further spread of rumors. To address these limitations, this paper proposes a novel method that integrates both temporal and spatial features while leveraging Large Language Models (LLMs) to automatically generate explanations for the detection results. Our method constructs a dynamic graph model to represent the evolving, tree-like propagation structure of rumors across different time periods. Spatial features are extracted using a Graph Convolutional Network, which captures the interactions and relationships between entities within the rumor network. Temporal features are extracted using a Recurrent Neural Network, which accounts for the dynamics of rumor spread over time. To automatically generate explanations, we utilize Llama-3-8B, a large language model, to provide clear and contextually relevant rationales for the detected rumors. We evaluate our method on two real-world datasets and demonstrate that it outperforms current state-of-the-art techniques, achieving superior detection accuracy while also offering the added capability of automatically generating interpretable and convincing explanations. Our results highlight the effectiveness of combining temporal and spatial features, along with LLMs, for improving rumor detection and understanding.