Dialogue Response Generation for Text-Based Dialogue Systems with Emotion Regulation

Abstract

Dialogue systems that handle emotions have been shown to improve user satisfaction and increase positive interactions, and are expected to play a role as a digital partner that accompanies humans. In order for dialogue systems to recognize user emotions and express their own emotions, emotion regulation methods that select appropriate emotions from the context of the dialogue are indispensable. In this paper, we propose a text-based dialogue framework that performs emotion regulation, which estimates appropriate emotions during response. Our proposed method estimates appropriate emotions by considering the semantic and emotional context of the user’s text, and generates dialogue responses with emotional expressions using a neural network. The effectiveness of the proposed method was demonstrated through automatic evaluation using an evaluation metrics for the quality of dialogue generation and human evaluation by 100 subjects collected through a crowdsourcing.