Research on Chinese Text Classification Based on Improved RNN

The traditional RNN model uses the hidden state of the time step for information storage in dealing with the Chinese text classification problem, which causes the subsequent vocabulary to have a greater impact on the classification results, and the classification accuracy is thus affected. To improve the model structure, two improvement methods are proposed in this paper. First, the word vector embedding is performed and then the BiLSTM model is connected to extract the global text features in both directions. secondly, the Attention calculation module is added to the traditional structure for deep feature interaction, or the pooling module of CNN is connected to perform feature dimension reduction. The classification efficiency and accuracy of the model are improved. In the experiment, the classification accuracy rate on the Chinese news dataset reached a maximum of 91.08%, which was 1.73% higher than the highest value of the benchmark model, which effectively improved the performance of the model for Chinese text classification.

I. Introduction

In recent years, text data has grown rapidly through the Internet, and more and more text data have been accumulated. The acquisition of text data is very extensive, and many paths, including network data, social media, and user comments, can provide a good source of information for text data. There are so many valuable information hidden in this information. How to efficiently mine and utilize this information is the task of natural language processing [1]. Text classification technology, including its sub-task sentiment analysis, is also a key part of it. However, the biggest difference between text data and other data types lies in its unstructured characteristics, that is, the internal connection of each text is hidden in the semantic information, and the internal semantic information cannot be completely obtained from the external structure. Therefore, the key to solving the problem of text classification is how to obtain semantic information effectively. Although the field of text representation has developed rapidly, there are still some problems in the application in the field of Chinese text. At present, almost all text representation models are constructed by relying on the English context, while the semantic richness of Chinese is stronger than that of English, with more complex grammar and more profound meanings. Therefore, it is difficult for these models to achieve the same excellence in the Chinese field. Effect. It is not enough to solve the Chinese text classification problem only by relying on the text representation model. It is usually necessary to integrate the text representation model with other deep learning models to work together to better achieve Chinese text classification. Based on the previous achievements, this paper improves the traditional RNN algorithm for the problem of Chinese text classification. Experiments show that the improved RNN model can effectively extract Chinese semantic features and realize the effective classification of Chinese texts.