CEGI: Commonsense-Enhanced Generative Inference for Event Prediction

In the field of event prediction (EP), predicting subsequent events for a given contextual event is an important and challenging task, which requires inferring the correlation between the context and the candidate event. Current research focuses on improving event prediction using external knowledge and pretrained language models, but faces the problems of sparse correlation knowledge and biased lateralization of generative inference. Thus, this paper introduces a novel approach termed Commonsense-Enhanced Generative Inference (CEGI), which enhances generative inference through commonsense relations between events. The CEGI approach constructs event relations through a commonsense knowledge base and updates the event representations using graph neural networks, and then performs event prediction through a generative approach. Experimental results on the multi-choice narrative cloze (MCNC) task demonstrate that our method outperforms state-of-the-art baseline methods and effectively enhances the accuracy of generative reasoning.