Research on Sensitive Text Classification Based on Knowledge Base and Hybrid Network

Aiming at the problems that convolution neural network is not sufficient to extract text features, it is difficult to capture long text structure information and sentence semantic relationship, and neural network can only extract the surface features of text, and it is difficult to obtain the implicit features of sentences when classifying text, this paper proposes a hybrid neural network text classification method based on a knowledge base. By extracting keywords from the category text, the keyword weight is calculated according to the co-occurrence relationship, and the category keyword knowledge base is constructed. It integrates CNN and bilstm, introduces attention mechanism, enhances the ability to extract local and temporal features of text, and improves the accuracy of text classification in combination with knowledge base information. The experimental results on the CNews dataset show that compared with TextCNN, TextRCNN, BILSTM and Bert models, this model can effectively improve the accuracy of text classification.