Make Information Diffusion Explainable: LLM-based Causal Framework for Diffusion Prediction

Information diffusion prediction, which aims to forecast the future infected users during the information spreading process on social platforms, is a challenging and critical task for public opinion analysis. With the development of social platforms, mass communication has become increasingly widespread. However, most existing methods based on GNNs and sequence models mainly focus on structural and temporal patterns in social networks, suffering from spurious diffusion connections and insufficient information for diffusion analysis. We leverage the strong reasoning capabilities of LLMs and develop an LL M -based causal framework for d iffusion i nf l uence d erivation, named MILD. By comprehensively integrating four key factors of social diffusion—i.e., connections, active timelines, user profiles, and comments—MILD causally infers authentic diffusion links to construct a diffusion influence graph, G I . To validate the quality and reliability of our constructed graph G I , we propose a newly designed set of evaluation metrics for diffusion prediction. In experiments, MILD provides a reliable information diffusion structure that achieves an absolute improvement of 12% over the social network structure and achieves state-of-the-art performance in diffusion prediction. MILD is expected to contribute to higher-quality, more explainable, and more trustworthy public opinion analysis. The code and data are available at: https://github.com/Shang-hub/ MILD-Official-Implementation .