基于BERT-CW的公共卫生事件情绪识别模型研究

Research on Emotion Recognition Model of Public Health Events Based on BERT-CW

针对情绪识别任务中,单一存在的模型存在片面性,无法充分提取语义特征等问题,本文提出了一种领域情感词典与字词特征融合相结合的文本分类方法。首先构建并扩展有关于“传染性疾病”事件的领域情感词典,其次融合文本的字向量特征和词向量特征,最后将BERT模型应用于“传染性疾病”事件微博文本分类任务中。实验结果显示,相较于其它神经网络模型,BERT-CW(字词融合)模型的精确率、召回率和F1值各项评价指标的表现更好;相比于字划分或词划分的BERT-C模型和BERT-W模型,BERT-CW模型的可靠性更高,实验结果在微博用户评论数据集的网络情绪识别任务上准确率达到了94.59%,F1值达到了94.08%,证实了此模型的有效性。

In order to solve the problems of one-sidedness and inadequacy of semantic features in emotion recognition tasks,a text classification method combining domain sentiment dictionary and word feature fusion is proposed in this paper.Firstly,a domain sentiment dictionary about“infectious disease”event is constructed and extended.Secondly,word vector features and word vector features of text are integrated.Finally,Burt model is applied to the classification task of“infectious disease”event microblog post.The experimental results show that compared with other neural network models,the accuracy rate,recall rate and F1 value of BERT-CW model perform better.Compared with the BERT-C model and the BERT-W model of word division or word division,the BERT-CW model has higher reliability.The experimental results show that the accuracy of the Internet emotion recognition task in the microblog user comment data set reaches 94.59%,and the F1 value reaches 94.08%,which confirms the validity of this model.