EMNLPNov, 2022

UniMSE: 统一多模态情感分析与情感识别

TL;DR本文提出一种多模态情感知识共享框架(UniMSE),用于统一多模态情感分析(MSA)和对话中情绪识别(ERC)任务以 better capture the difference and consistency between sentiments and emotions by performing modality fusion at the syntactic and semantic levels and introducing contrastive learning between modalities and samples. 通过在四个公共测试数据集上的实验,展示了该方法的有效性并与最先进的方法进行了一致的改进。