Knowing What and Why: Causal emotion entailment for emotion recognition in conversations

Abstract

The clues for eliciting emotion deserve attention in the realm of Emotion Recognition in Conversations (ERC). In an ideal dialog system, comprehending emotions alone is insufficient, and underlying the causes of emotion is also imperative. However, previous research overlooked the integration of causal emotion entailment for a prolonged period. Therefore, an emotion-cause hybrid framework that utilizes causal emotion entailment (CEE) is proposed to promote the ERC task. Specifically, the presented method integrates the information of the cause clause extracted through the CEE module that triggers emotions into the utterance representations obtained by the ERC model. Moreover, a Bidirectional Reasoning Network (BRN) is designed to extract emotional cues to simulate human complex emotional cognitive behavior. Experimental results demonstrate that our framework achieves a new state-of-the-art performance on different datasets, indicating that the proposed framework can improve the model’s ability to emotion understanding.