时 间:2024-11-05 13:30-14:30
地 点:线下 李达三楼104室;线上 腾讯507326317/722722
题 目:Large Language Model as an Empathetic Mind Simulator: A Novel Deep Learning Method for Conversation-based Empathy Need Identification
主讲人: 陈刚 信息管理与商业智能系 青年副研究员
内容摘要:
Enhancing the empathy capabilities of large language models (LLMs) and human agents signifies immense commercial value in today’s era of AI-generated content. However, identifying users’ empathy needs based on their conversations with an AI or human agent is non-trivial due to the difficulty in gauging their subjective needs and understanding their minds beyond the conversations. To tackle this challenge, we propose to enhance empathy need identification through user mind simulation by leveraging an LLM’s understanding and reasoning capabilities. Inspired by the spreading activation theory and the appraisal theory, we design a novel empathetic mind simulation-empowered deep learning (EMS-DL) method, which identifies users’ cognitive needs through in-conversation need spreading and activation with a cognitive mind simulator, recognizes users’ emotional needs through emotion contagion and activation with an emotional mind simulator, and optimizes its empathy need identification capability by drawing guidance from human agent experience with a generative contrastive learning-based optimization scheme. Empirical evaluation based on two (patient-doctor and e-commerce customer-sales agent) conversation datasets demonstrates the superiority of the empathy needs identified by EMS-DL over those by state-of-the-art alternatives, in terms of both empathy needs-empowered satisfactory response generation and empathy needs-based conversational satisfaction prediction. Explanatory analysis renders insights into users’ need spreading paths and emotion contagion patterns, as well as how the spread needs and contagious emotions enhance empathy need identification
信息管理与商业智能系
2024-10-28
活动讲座
新闻动态
微信头条
招生咨询
媒体视角
瞰见云课堂