Clause Level Attention Network for Emotion Cause Pair Extraction

Abstract

The Emotion-Cause Pair Extraction task aims to identify one or more pairs of emotion-causes in a document, and earlier work has focused on the implicit interaction between ECPE and two auxiliary tasks Emotion Extraction and Cause Extraction. However, existing models cannot capture the complex connection between emotion and cause. To address this problem, we propose an end-to-end model. The model utilizes properties of graph attention networks (GATs) to model the bidirectional dependencies between emotion and cause clauses, and experimental results on a benchmark emotion cause corpus verify the feasibility of our approach.