基于BERT-BiLSTM-CNN模型的新闻文本分类研究

Research on News Text Classification Based on BERT-BiLSTM-CNN

针对双向长短期记忆网络(BiLSTM)没有考虑到局部关键信息对文本分类的影响,以及卷积神经网络(TextCNN)无法捕获文本的长远距离的特征信息等问题,文章提出了一种基于BERT-BiLSTM-CNN混合神经网络模型的新闻文本分类的方法。为了进一步增强文本表示和提高新闻文本分类的效果,首先使用BERT预训练模型对文本进行词嵌入映射,其次利用BiLSTM-CNN模型进一步提取文本上下文和局部关键特征,最后对新闻文本进行分类;并在THUCNews数据上进行对比实验,实验结果表明,BERT-BiLSTM-CNN模型的文本分类效果优于Transformer、TextRNN、TextCNN等深度学习模型。

Aiming at the problems that Bi-directional Long Short-Term Memory(BiLSTM)network does not consider the influence of local key information on text classification,and that TextCNN cannot capture the long-term feature information of text,this paper proposes a method of news text classification based on BERT-BiLSTM-CNN hybrid neural network model.In order to further enhance text representation and improve the effectiveness of news text classification,firstly the BERT(Bidirectional Encoder Representation from Transformers)pre-training model is used to perform word embedding mapping on the text.Secondly,the BiLSTM-CNN model is used to further extract text context and local key features.Finally,the news text is classified and comparative experiments are conducted on THUCNews data.The experimental results shows that the text classification effect of BERT-BiLSTM-CNN model is better than that of Transformer,TextRNN,TextCNN and other deep learning models.