Emotion Cause Extraction by Combining Intra-clause Sentiment-enhanced Attention and Inter-clause Consistency Interaction

Emotion cause extraction aims at identifying causes of the given emotion expression in a document. Many approaches have been proposed to solve this task, including rule-based approaches, traditional machine learning methods and deep neural networks. However, these approaches focus more on text content and suffer from the problem of the information deficiency issue. In this work, we exploit external sentiment knowledge, intra-clause syntactic dependency and inter-clause consistency to alleviate the problem. We design a sentiment-enhanced attention mechanism to fuse sentiment knowledge and syntactic dependency, and we obtain consistency information through the interaction between clauses on the same side of emotion expression. We finally achieve the best performance among 10 compared methods.