结合BERT词嵌入和双向循环卷积神经网络的新闻文本分类研究

A Study of News Text Classification Combining BERT Word Embedding and Bidirectional Recurrent Convolutional Neural Network

针对数字信息时代网络舆情爆发的复杂性及不可控性,提出一种融合BERT、TEXTRCNN、BILSTM-CRF的新闻文本分类模型,致力于提高新闻文本分类的准确率,通过采用BERT词嵌入技术高效获得句子语义特征,利用TEXTRCNN双向递归的结构以及BILSTM-CRF模型的运用来解决序列标注问题,综合考虑上下文捕捉、词嵌入、文本特征等因素,提高对新闻识别的准确性。实验证明,使用该融合模型对新闻文本分类的准确率达到0.9551,且具有较好的泛化能力,能够更好地帮助有关部门及时处理突发舆情和失控事件。

Aiming at the complexity and uncontrollability of the outbreak of network public opinion in the digital information age,a news text classification model integrating BBRT,TEXTRCNN and BILSTM-CRF is proposed,Committed to improving the accuracy of news text classification.The semantic features of sentences are efficiently obtained by using BERT word embedding technology,and the structure of TEXTRCNN two-way recursion is used.And the use of the BILSTMCRF model to solve the sequence annotation problem,comprehensively considering the factors such as context capture,word embedding,text features,etc.,to improve the accuracy of news recognition.Experiments have shown that the accuracy rate of news text classification using the fusion model is 0.9551,and it has good generalization ability,which can better help relevant departments deal with sudden public opinion and out-of-control events in a timely manner.