Multi-Task Learning-based Research on the Problem of Conversational Emotion Recognition
Huilin Mou, Yu Wang, Guangyu Lei · 2023
Emotion Recognition in Conversations (ERC) can determine the emotion state of a conversation by analyzing text or speech information in the conversations, which has led to its widespread interest. Many approaches also exist dedicated to contextual understanding through deep learning. The paper proposes the novel Multi-Task Learning Network for Emotion Recognition in Conversations (DialogueMTLN). Two networks are built separately, and the two tasks proceed in parallel using a multi-task learning approach. The respective loss functions are weighted and summed by an adaptive method and then back-propagated to train the network, establishing the critical process of emotion cue inference, achieving a complete understanding of the conversational context, and finally accomplishing an accurate classification of the emotion in the conversations. In extensive experiments, the proposed model is shown to be effective and superior on two public benchmark datasets.