融合时间序列和知识增强的双图融合对话情感识别

Dialogue emotion recognition based on dual-graph fusion integrating time series and knowledge enhancement

针对对话中的情感识别主要集中在对话的上下文建模以及说话者之间的依赖关系建模,而忽略了对话的时间序列特征问题,通过时间编码分析会话情感在时间序列上的变化,更好地捕捉情感变化的趋势;引入外部知识作为辅助信息,并设计3个步骤进行知识选择,为对话有效选择相关的知识信息,为移情反应提供一定的指导;设计了一个双图融合模块进行对话信息的交互,以此提升模型的整体性能。在两个公开数据集上实验表明,提出的模型与基线模型相比,性能指标更优,效果更好。

Current dialogue emotion recognition research primarily focuses on modeling the dialogue context and the dependency relationships between speakers,while often overlooking the time series characteristics of dialogue.By using time encoding to analyze the emotional changes in conversations over time,we can better capture the trends in emotional shifts.This work introduces external knowledge as auxiliary information and designs a three-step process for knowledge selection,enabling the model to effectively choose relevant knowledge for the dialogue,thereby providing guidance for empathetic responses.Additionally,a dual-graph fusion module is designed to facilitate interaction between dialogue information,improving the overall performance of the model.Experiments conducted on two public datasets show that the proposed model outperforms baseline models in terms of performance metrics and overall effectiveness.