【摘要】 提出一个多轮对话情绪识别模型，首先使用强大的文本特征提取能力的RoBERTa预训练进行文本编码；然后将多轮文本信息按时间步输入到GRU结构中，解决了过去模型没有同时考虑一句对话中的语义相关性和多轮对话之间的时序关系问题。通过在4个数据集上的实验结果表明，文中模型能够有效提高对话生成中情感分析的性能，F₁值最高提高了1.24个百分点。更多还原

【Abstract】 This paper proposes a multi-turn dialogue emotion recognition model, which first uses RoBERTa pre-training with more powerful text feature extraction capabilities for text encoding, and then inputs multi-turn text information into the GRU structure according to the time step. This solves the problem that previous models did not consider both the semantic relevance in a sentence and the temporal relationship between multi-turn dialogues simultaneously. Experimental results on four datasets show that the proposed model can effectively improve the performance of sentiment analysis in dialogue generation, with the highest F₁ value increasing by 1.24 percentage points.更多还原