Emotion-Cause Pair Extraction Based on Dependency-injected Dual-MRC

Emotion-Cause Pair Extraction (ECPE) aims to extract emotion clauses and their corresponding cause clauses from text. This technique is helpful for understanding the emotional states conveyed in text and the causes behind them, thereby providing support for applications such as emotion intervention and public opinion guidance. We propose an ECPE method by incorporating dependency parsing into a dual Machine Reading Comprehension (MRC) model, enabling joint optimization from emotion to cause and back from cause to emotion. We transform the dependency tree into clause-level matrices and employ graph attention networks to handle relationships between clauses. Experimental results demonstrate the effectiveness of incorporating dependency parsing and utilizing the dual-MRC framework.