Leveraging Causal Convolutional Neural Networks and Knowledge Matching for Enhanced Event Causality Identification

Event Causality Identification (ECI) is a vital task in the field of natural language processing, focusing on identifying causal relationships between pairs of events in texts. This task is currently receiving significant attention, but some achievements and the existing methods are still not effective enough. There are two issues: (1) Prior efforts primarily concentrated on external knowledge processing and event pairs, neglecting the utilization of causal word cues between events. (2) While ConceptNet incorporates extensive external knowledge, not all of it is effectively utilized by the models. Our paper introduces a novel framework combining convolution and knowledge matching, with two core modules to address the above problems. Specifically, to extract causal words between events, we use convolutional neural network (CNN) to capture the relationship between the pairs. For the causal index, we select the causal indicator from FrameNet to initialize the CNN filter. When introducing external knowledge, we use COMET model to generate causal knowledge, and design a causal matching algorithm to judge the relationship between events. Experimental results on two widely used datasets indicate that our method outperforms previous methods.