AECR framework boosts causal event prediction for rare, unseen scenarios

Event-centric AI systems rely on explicit causal knowledge for risk warning and decision support, but existing instance-level causal pairs struggle to generalize to low-frequency long-tail events and unseen combinations. To tackle this, Zheng, Chen, and Wang from the paper "Abstract Event Causal Rules: Induction and Application" introduce AECR (Abstract Event Causal Rule), a novel abstraction paradigm that transforms concrete cause-effect pairs into generalized abstract causal logic while preserving their intrinsic causal relationships. The approach uses a multi-agent Concrete-to-Abstract Causal Induction (CACI) system coupled with similarity-constrained clustering to distill trustworthy AECRs from noisy raw data, resulting in two complete AECR knowledge bases.

To validate practical utility, the authors propose AR-GCAE (Abstract Rule-Guided Causal Attention Encoder), which injects retrieved AECRs into the causality Graph Event Prediction (CGEP) benchmark via rule-guided attention layers and gated representation fusion. Quantitative results show that applying AECRs substantially strengthens generalization capacity for event causal reasoning and brings consistent performance improvements, with the most prominent gains observed on rare and unseen event samples. This work addresses a critical limitation in event prediction and opens a new direction for relation-level causal abstraction in AI analytics.