Research on News Text Classification Method Based on Hybrid Neural Networks

With the rapid advancement of Internet technologies, news websites and online media platforms have become crucial components of the news dissemination landscape. Online news has also become one of the primary sources for people to obtain information. To meet the needs of numerous internet news readers and ensure that content providers can efficiently deliver personalized news recommendations, the effective control and utilization of internet news information have become essential. In this context, a news text classification method based on hybrid neural networks has emerged. Addressing the limitations of E-TF-IDF in neglecting positional relationships, this paper proposes a function that incorporates positional variables, denoted as P, based on readers' habits and the normal distribution. This study integrates the advantages of Bert, CNN, and BiLSTM. First, the preprocessed data is fed through the E-TF-IDF-P algorithm into the Bert model. Then, local features are extracted via TextCNN, and global features are captured via BiLSTM. These features are fused and input into the softmax layer for text classification.