共情对话研究进展  

Research progress of empathetic dialogue system

作　　者：何俊[1] 饶方喜 周志豪 徐秋 He Jun;Rao Fangxi;Zhou Zhihao;Xu Qiu(College of Information&Engineering,Nanchang University,Nanchang 330031,China)

机构地区：[1]南昌大学信息工程学院,南昌330031

出　　处：《计算机应用研究》2024年第1期1-9,共9页Application Research of Computers

基　　金：国家自然科学基金资助项目(62066025)。

摘　　要：人工智能驱动的对话是当前研究热点,有着广泛的应用前景。但目前这类对话系统普遍缺乏情感交互能力,限制了其在心理关爱、抑郁症等精神障碍疾病方面的应用。如何让对话系统充分理解用户情绪并生成带有共情的回复是目前对话系统面临的主要挑战之一。首先介绍了共情对话研究中情感感知和共情对话生成两大挑战,并分别调研归纳了相关研究方法。情感感知任务大致可分为基于规则、基于机器学习和基于深度学习三类方法,共情对话生成大致可分为基于检索和基于动态生成两类方法。接着介绍了共情对话的最新发展动向,并总结了共情对话数据集、通用对话数据集和多模态数据集的特点和链接,归纳了当前共情对话研究中不同的评估方法便于后续研究。最后对共情对话的研究工作进行了总结和展望。Artificial intelligence-driven dialogue is a current research hotspot with a wide range of promising applications.However,such dialogue systems currently generally lack emotional interaction capabilities,limiting their application in psychological care,depression and other mental disorders.How to make dialogue systems fully understand users’emotions and generate responses with empathy is one of the main challenges facing dialogue systems today.This paper first introduced two major challenges in empathic dialogue research:emotion perception and empathic dialogue generation,and summarised the relevant research methods in separate studies.Emotion perception tasks could be broadly classified into rule-based,machine-learning and deep-learning approaches,while empathic dialogue generation could be broadly classified into retrieval-based and dynamic generation-based approaches.Then this paper introduced the latest developments in empathic dialogue,and summarised the features and links of empathic dialogue datasets,generic dialogue datasets and multimodal datasets,and summarised different evaluation methods in current empathic dialogue research to facilitate subsequent research.Finally,this paper presented a summary and outlook of the research work on empathic dialogue.

关 键 词：共情对话 情感感知 对话生成 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]