摘要
利用预训练模型对中文文本进行情感分析是目前的主流方式,K-BERT模型的提出克服了BERT模型不具备背景知识的问题。本文通过在K-BERT的基础上引入双向长短时记忆网络,提出了KB-BERT情感分析优化模型。首先,通过K-BERT预训练模型,对输入的内容进行背景丰富,获取包含背景知识的语义特征向量;其次,利用长短时记忆网络提取上下文的相关特征,进行文本情感分析。实验结果表明,使用KB-BERT的准确率优于K-BERT,在Book_review和Weibo两个数据集上的准确率,分别达到了87.97%和98.33%。
Using pretrained model to perform sentiment analysis on Chinese text is the current mainstream way. The proposal of the K-BERT model overcomes the problem that the BERT model does not have background knowledge. By introducing a two-way LSTM network on the basis of K-BERT, the KB-BERT sentiment analysis optimization model is proposed. Through the K-BERT pretrained model, the input content is enriched in the background, and the feature vector containing the contextual semantic information is obtained. Then LSTM is used to extract the relevant features of the context, and finally the text sentiment analysis is performed. The experimental results show that the effect of using KB-BERT is better than that of K-BERT, and the accuracy on the two data 87.97% and 98.33%, respectively.
作者
王桂江
黄润才
WANG Guijiang;HUANG Runcai(School of Electronic and Electrical Engineering,Shanghai University of Engineering Science,Shanghai 201620,China)
出处
《智能计算机与应用》
2022年第7期35-39,共5页
Intelligent Computer and Applications
