计算机应用研究2025，Vol.42Issue(7)：1964-1971,8.DOI:10.19734/j.issn.1001-3695.2024.12.0486
基于大语言模型的多任务生成式重构对话情绪识别
Multi-task generative emotion recognition in conversation based on large language models
摘要
Abstract
Emotion recognition in conversation(ERC)is a key task in dialogue systems research.However,existing models often suffer from overfitting to specific datasets and dialogue patterns due to the complexity of pipeline design,which limits their generalization ability.To address this issue,this study proposed a multi-task generative emotion recognition in conversation(M-GERC)model based on large language models.The model introduced two auxiliary tasks based on pre-trained large language models:speaker identification and topic-based emotion prediction.The speaker identification task aimed to implicitly model the relationships between conversational roles,helping the model better understand emotional exchanges between different partici-pants.The topic-based emotion prediction task predicted the global theme of the conversation,capturing the potential connec-tion between topics and emotions,thus improving emotion recognition accuracy by incorporating contextual information.Addi-tionally,M-GERC introduced a knowledge retrieval module that retrieved domain-specific knowledge and integrated external knowledge to further enhance the model's understanding of context.Experimental results show that M-GERC significantly out-performs existing mainstream ERC models,achieving W-F1 improvements of 3.1％,4.3％and 3.7％on the DailyDialog,MELD and EmoryNLP datasets,respectively.关键词
对话情绪识别/大语言模型/主题/外部知识Key words
emotion recognition in conversation/large language models/topic/external knowledge分类
信息技术与安全科学引用本文复制引用
龙禹辰,勾智楠,陈宇欣,秦乐..基于大语言模型的多任务生成式重构对话情绪识别[J].计算机应用研究,2025,42(7):1964-1971,8.基金项目
河北省自然科学基金资助项目(F2023207003) （F2023207003）
河北经贸大学科学研究与发展计划资助项目(2024YB23) （2024YB23）
河北经贸大学教学研究项目(2024JYQ09) （2024JYQ09）
