LE²C: LLM-Enhanced Event Evolutionary Graph for Explainable Classification

In the development of intelligent systems, multi-label event classification plays a vital role in enabling accurate decision-making across diverse scenarios such as incident response, customer service, and urban management. Although existing graph-based approaches for multi-label classification have shown potential, they struggle to model directed label dependencies and lack explainability, resulting in black-box decision-making processes. To address these issues, we propose a novel LLM-enhanced Event Evolutionary graph for Explainable Classification (LE²C) method. Specifically, we first leverage the powerful semantic learning capabilities of Large Language Models to construct an event evolutionary graph that models event dynamics. Furthermore, we introduce a co-occurrence probability matrix to enhance the expressivity and explainability of the graph, guiding explainable classification. Extensive experiments on two large real-world event classification tasks demonstrate the efficiency, effectiveness, and explainability of LE²C. The code is available at https://github.com/NinaLiangjy/LE2C.