An Emotion Guidance Decision-Making Model Based on Dialogue Context
Zhiqiang Ma, Wenchao Jia, Baoxiang Du, Chunyu Wang · 2021
Emotional dialogues are mainly for small chats, companionship, emotional comfort and other scenarios. In most scenarios, we hope to guide the user's emotions in a positive direction in the dialogue. However, the lack of appropriate emotion guidance mechanisms may lead to unexpected emotions in dialogue response, which may hurt the user's feelings and reduce the user's willingness to chat. Current dialogue models have achieved great empirical success in emotional expression and controllability of responses, but there is not yet an effort in user emotion guidance decision-making. In this paper, inspired by time-series models and the dynamic characteristics of emotional changes in dialogue, we propose an emotion guidance decision-making model with an emotion guidance decision-making cell to decide a proper emotion for the current response according to the specified sentiment. Experimental results of automatic and human evaluations demonstrate that the proposed emotion guidance decision-making model can effectively decide the emotions that are consistent with the target sentiment.