Enhancing Official Policy Document Classification with BERT and Keyword Augmentation

Official policy document (OPD) classification is a kind of special text classification. Good policy classification can bring great convenience to the people. However, on many local government websites, such policies are not well separated, which brings trouble to enterprises and individuals. For this reason, the job of policy classification becomes very necessary. Although many government officers have used different models to classify relevant policies, there are many cases in which the actual effect is not ideal. There are two main reasons for this situation. First, Chinese text classification has its own unique difficulties compared with English text classification. The second is the lack of data. The authors in this paper propose a method to enhance policy text classification using BERT. The proposed method uses keyword extraction and data augmentation techniques to improve the accuracy of the policy classification model. Specifically, the method extracts keywords from OPD using Jieba's keyword extraction module and utilizes data augmentation to increase the size and diversity of the training dataset. The experimental results show that compared with other traditional classification models, the proposed method achieves better performance. Overall, this work demonstrates the potential of BERT-based methods for policy text classification and provides insights into the effective use of keyword extraction and data augmentation in natural language processing tasks.