Leveraging LLMs for Diffusion Prediction in Social Networks: A Fused Attention-aware Model

Information diffusion prediction, which aims to forecast future infected users during the information spreading process in social networks, is a challenging and critical task for applications such as social recommendation and fake news detection. Most existing methods primarily leverage GNNs and RNNs to capture structural and temporal patterns on social networks. However, the rich information contained in user profiles and post content is neglected, and the diffusion influence between users is challenging to model. To address these limitations, this paper proposes a fused attention-aware diffusion prediction method based on our fine-tuned Diffusion-LLM to improve the accuracy and interpretability of prediction results. The fused attention mechanism incorporates the diffusion influence analyzed by Diffusion-LLM and the user similarity matrix derived from a Transformer model based on user profiles. Preliminary experiments are conducted to verify the feasibility of our Diffusion-LLM.