Multi-Granularity Context-Aware and Iterative Graph Optimization for Document-Level Event Causality Identification

Event causality identification (ECI) aims to detect causal relationships between events in textual contexts. Compared to sentence-level ECI which focuses on an individual sentence. Document-level ECI (DECI) presents significantly greater challenges of processing long texts and cross-sentence events than sentence-level ECI (SECI). Current document-level ECI approaches face two key challenges: (1) learning high-quality context-aware event representation to capture event-specific causal semantic and (2) effectively integrating structural and semantic information for causal reasoning. To address these limitations, we propose MAIGO (multi-granularity context-aware and iterative graph optimization), a novel DECI framework that revolutionizes DECI through multi-granularity attention mechanisms for context-aware event representation and dynamic event causality graph structure iterative optimization via topology-embedding mutual optimization. MAIGO enhances event embeddings through sentence-level, document-level, and graph-level attention, capturing event-event, event-sentence, and sentence-sentence interactions to produce better representation for events. The framework further constructs an event causality graph (ECG) and jointly optimizes both the ECG structure and event embeddings via masked attention in an iterative optimization process. Extensive experiments on two benchmark datasets and a multilingual dataset demonstrate MAIGO’s effectiveness, achieving improvements of 4.1% (causality existence identification) and 7.3% (causality direction identification) in inter-sentence ECI compared to state-of-the-art models. Further analysis confirms the advantages of our multi-granularity context-aware enhanced representation strategy and iterative graph optimization mechanism.