A New Component of Interactive Multi-task Network Model for Emotion-Cause Pair Extraction

Abstract

The task emotional-cause pair extraction (ECPE) is proposed based on the emotion cause extraction (ECE) task. The ECPE task proposes a new two-step frame method: first step is to extract emotion and cause information separately, then the second step is to match the emotional cause pair. Compared to the ECE task, this new research no longer relies on manual annotation of documents. However, the Inter-EC model component for the ECPE task does not fully consider the relevance of sentimental clause and causal clause, and the available context is limited. In this paper, we propose a new extraction component with self-attention for the Inter-EC model, optimizing the interactive multi-task network model. Our approach achieved a good recall score on the ECPE dataset, and the F1 value is slightly better than the baseline.