Event Causality Identification(ECI)aims to identify causal relationships between events in unstructured text.This task is extremely challenging,as causal relationships are often expressed through implicit associations between events.Existing meth-ods typically involve directly modeling text with pre-trained language models to capture this association,which struggles to grasp the semantic relationships of event pairs over long text spans.As the distance between event pairs in the text increases,the model's abil-ity to capture their semantic relationship weakens.By modeling the semantic structure of the text,implicit causal associations be-tween events are studied.The use of graph neural network-based event aggregators integrates structural information of event pairs,and the syntactic information of the text is used to semantically enhance the representations of event pairs extracted by the model.Ex-perimental results on two widely used datasets show that the model's performance significantly outperforms baseline methods.
维普
万方数据