【摘要】 为了减少文本情感分析方法中对分词准确性的依赖程度,同时,也为了解决一词多义的问题,提出了一种基于BERT和双向长短时记忆网络(BiLSTM)的文本情感分类模型.首先,该模型采用BERT(Bidirectional Encode,Representation from Transformers,BERT)预训练的字向量替代传统方式训练的词向量,然后利用BiLSTM对文本上下文进行关系特征提取,最后再利用Softmax分类器获得文本所属情感类别.实验证明,与LSTM、BiLSTM、TextCNN和BERTLSTM模型相比,该模型在综合评价指标F1上分别提高了6.78%、6.74%、2.52%和1.01%;测试集正确率分别提升了6.66%、6.31%、1.95%和0.85%.更多还原

【Abstract】 In order to reduce the dependence on the accuracy of word segmentation in text sentiment analysis,and to solve the problem of polysemy,a text sentiment classification model based on Bert and BiLSTM is proposed.Firstly,Bert pre-trained character vector is used to replace the word vector,then Bi-LSTM is used to extract contextual features,and finally Softmax classifier is used to get the emotional category of the text.Compared with LSTM,BiLSTM,TextCNN and Bert-LSTM models,the model was improved by 6.78%,6.74%,2.52%and 1.01%in the comprehensive evaluation index F1,respectively.The accuracy of test set was improved by 6.66%,6.31%,1.95%and 0.85%,respectively.更多还原