Think Twice: A Human-like Two-stage Conversational Agent for Emotional Response Generation

TL;DR

This paper introduces a two-stage conversational agent inspired by human 'think twice' behavior, which generates emotionally appropriate responses while maintaining semantic relevance, overcoming limitations of joint emotion-semantic models.

Contribution

The paper proposes a novel two-stage framework for emotional response generation that separates semantic and emotional modeling, improving response quality and emotional appropriateness.

Findings

Outperforms comparison models in emotion generation

Maintains semantic relevance in responses

Effective on DailyDialog and EmpatheticDialogues datasets

Abstract

Towards human-like dialogue systems, current emotional dialogue approaches jointly model emotion and semantics with a unified neural network. This strategy tends to generate safe responses due to the mutual restriction between emotion and semantics, and requires rare emotion-annotated large-scale dialogue corpus. Inspired by the "think twice" behavior in human dialogue, we propose a two-stage conversational agent for the generation of emotional dialogue. Firstly, a dialogue model trained without the emotion-annotated dialogue corpus generates a prototype response that meets the contextual semantics. Secondly, the first-stage prototype is modified by a controllable emotion refiner with the empathy hypothesis. Experimental results on the DailyDialog and EmpatheticDialogues datasets demonstrate that the proposed conversational outperforms the comparison models in emotion generation and maintains the semantic performance in automatic and human evaluations.