Emotion-cause pair extraction based on interactive attention

In recent years, a new fine-grained task has been proposed in the field of sentiment analysis, the emotion-cause pair extraction (ECPE) task, whose purpose is to extract all emotions and their causes from a document. Most of existing methods produce effective emotion-cause pairs by filtering all possible pairs. However, this types of methods ignore the relationship between emotion clauses and cause clauses when learning the representations of emotions and causes clauses. In order to solve the above problem, we propose an end-to-end framework, which uses interactive attention and its fusion mechanism to learn the relationship between emotions and causes, and then pair them. Experimental results on quasi-base corpus shows our proposed method outperform the state-of-the-art baseline.