情感导向对话回复生成模型

Emotion Guided Dialogue Response Generation Model

情感对话生成模型在生成回复中并未考虑对话双方的情感变化,导致生成回复的情感存在不确定性,降低了用户的对话意愿。该文基于情感诱发相关研究提出了一种情感导向回复生成模型EGRGM,模型添加了情感导向门,控制机器情感状态合理生成,逐步引导用户情感向目标情感自然转化。实验以NLPCC2018的开放领域多轮对话语料为基础,基于GPT-2构建情感导向对话回复生成实验系统,开展了情感引导、逻辑关联以及对话持续等实验。实验结果表明,模型在情感引导方面与最优模型相比提高了13%,能够根据对话历史和指定的引导目标情感极性,控制用户回复生成的情感,实现对用户情感的引导。

The emotion dialogue generation model does not consider the change of emotions of both parties in the dialogue in generating responses,which leads to uncertainty in the emotions of the generated responses and reduces the use's willingness to dialogue.In this paper,an emotion-guided reply generation model EGRGM is proposed based on emotion-evoking related research,and the model adds an emotion-guided gate to control the rational generation of machine emotion states,gradually guiding the user's emotions towards the natural transformation of the target emotion.Theexperiment dataset was constructed with the open-domain multi-round dialogue corpus from NLPCC2018,experiments on emotion guidance,logical association and dialogue persistence are conducted.The experimental results show that the model improves the emotion guidance score by 13%compared with the optimal model,indicating that the model is able to control the generation of the user's emotional reply and realize the guidance of the user's emotion according to the conversation history and the specified target emotional polarity.