心理科学进展2025，Vol.33Issue(10)：1783-1793,11.DOI:10.3724/SP.J.1042.2025.1783
大语言模型的共情模拟:评估、提升与挑战
Empathy in large language models:Evaluation,enhancement,and challenges
摘要
Abstract
With advances in large language models(LLMs)in natural language generation and affective computing,their ability to simulate empathy has attracted increasing attention in domains such as psychological counseling,doctor-patient communication,and customer service.Empathy simulation in LLMs primarily manifests as cognitive rather than emotional empathy,encompassing emotion recognition,empathic response generation,and contextual adaptation.Current evaluation methods for empathic simulation in LLMs include human evaluation,automated evaluation,and task-based evaluation,each with its own advantages and limitations depending on the application context.Compared to human empathy,LLMs perform well in empathy generation tasks but still face fundamental limitations in emotional understanding.To further enhance the empathic capabilities of LLMs,techniques such as data augmentation,model architecture optimization,reinforcement learning,and prompt engineering can be employed.Meanwhile,ethical considerations and potential risks associated with the use of these models remain important concerns.关键词
大语言模型/共情模拟/共情评估/伦理问题Key words
large language models/empathy simulation/empathy evaluation/ethical issues分类
社会科学引用本文复制引用
周倩伊,蔡亚琦,张亚..大语言模型的共情模拟:评估、提升与挑战[J].心理科学进展,2025,33(10):1783-1793,11.基金项目
青少年心理健康与危机智能干预安徽省哲学社科重点实验室开放基金重大项目(SYS2024XXX)资助. （SYS2024XXX）
