The concept of empathy is vital in human-agent systems as it contributes to mutual understanding, problem-solving and sustained relationships. Despite the increasing adoption of conversational systems as one of the most significant events in the recent decade, the emotional aspects require considerable improvements, particularly in effectively displaying empathy. This paper provides a critical review of this rapidly growing field by examining the current advances in four dimensions: (i) conceptual empathy models and frameworks, (ii) the adopted empathy-related concepts, (iii) the datasets and algorithmic techniques developed, and (iv) the evaluation strategies. The review findings show that the most studies centred on the use of the EMPATHETICDIALOGUES dataset, and the text-based modality dominated research in this field. Moreover, studies have focused mainly on extracting features from the messages of both users and the conversational systems, with minimal emphasis on user modelling and profiling. For implementation in variegated real-world domain settings, we recommend that future studies address the gaps in detecting and authenticating emotions at the entity level, handling multimodal inputs, displaying more nuanced empathetic behaviours, and encompassing additional dialogue system features.

本研究通过五个评估维度研究人们在对话系统中尝试引入共情时所存在的关键点和未来机会，发现用户建模和对用户情感的检测和鉴别是需要进一步研究的核心问题；同时，处理多模式输入，显示更细腻的共情行为，以及包括其他对话系统特征也是需要关注的问题。

共情式对话系统：现有进展、差距和机遇综述