Emotion-Cause Pair Extraction Based on Structural and Semantic Heterogeneous Graph

In recent years, textual emotion-cause pair extraction (ECPE) has become a hot research topic today and has shown strong practical application value. Nevertheless, the majority of current research on ECPE is confined to single narrative text scenarios, with dialogue text scenarios that have more practical applications being largely overlooked. In this paper, we present an ECPD model based on a structural and semantic heterogeneous graph as well as label sequence prediction. The model constructs a structural and semantic heterogeneous graph to model conversations. It uses the properties of heterogeneous graphs to capture semantic information from multiple perspectives and extracts features of conversational texts using graph convolution operations. Trained on the RECCON dataset, the F1 of the ECPD model are all superior to those of the benchmark models in the literature, with the other evaluation metrics also located in the top two. This demonstrates the excellent performance of the ECPD model on the ECPE task of dialogue text.