Abstract Event Causal Rules Enhance Event Prediction Generalization

Event-centric analytical systems, crucial for risk warning and decision-making, often rely on explicit causal event knowledge. However, existing instance-level causal pairs struggle with generalization, particularly for low-frequency or unseen event combinations. This paper proposes Abstract Event Causal Rule (AECR) to address this limitation. AECR is a new relation-level causal abstraction paradigm that transforms concrete cause-effect pairs into generalized abstract causal logic while preserving their intrinsic causal relationships. The researchers designed a multi-agent Concrete-to-Abstract Causal Induction (CACI) system, combined with similarity-constrained clustering, to distill trustworthy AECRs from noisy raw causal data. This process led to the creation of two comprehensive AECR knowledge bases. To demonstrate the practical utility of this abstract causal knowledge, an Abstract Rule-Guided Causal Attention Encoder (AR-GCAE) was developed. This encoder injects the retrieved AECRs into the causality Graph Event Prediction (CGEP) benchmark task using rule-guided attention layers and gated representation fusion. Quantitative experiments showed that applying AECRs substantially strengthens the generalization capacity of event causal reasoning, leading to consistent performance improvements in event prediction, with the most notable gains observed for rare and unseen event samples.