School of Economics, Hebei University, Baoding, China.
Front Public Health. 2024 Feb 19;12:1264827. doi: 10.3389/fpubh.2024.1264827. eCollection 2024.
The application of health industry policies could be discovered more quickly and comprehensively through the automated identification of policy tools, which could provide references for the formulation, implementation, and optimization of subsequent policies in each province. This study applies the Bidirectional Encoder Representation from Transformer (BERT) model to identify policy tools automatically, utilizes Focal Loss to reduce the unbalance of a dataset, and analyzes the evolution of policy tools in each province, which contains time, space, and topic. The research demonstrates that the BERT model can improve the accuracy of classification, that supply and environment policy tools are more prevalent than demand tools, and that policy instruments are organized similarly in four major economic regions. Moreover, the policy's attention to topics related to healthcare, medicine, and pollution has gradually shifted to other topics, and the extent of policy attention continues to be concentrated on the health service industry, with less attention paid to the manufacturing industry from the keywords of the various topics.
通过自动化识别政策工具，可以更快、更全面地发现卫生行业政策的应用，为各省后续政策的制定、实施和优化提供参考。本研究应用双向转换器（Bidirectional Encoder Representation from Transformer，BERT）模型自动识别政策工具，利用焦点损失（Focal Loss）减少数据集的不平衡，并分析各省政策工具的演变，包括时间、空间和主题。研究表明，BERT 模型可以提高分类的准确性，供应和环境政策工具比需求工具更为普遍，四大经济区域的政策工具组织方式相似。此外，政策对与医疗保健、医学和污染相关的主题的关注逐渐转移到其他主题，政策关注度持续集中在卫生服务业，各主题的关键词表明对制造业的关注度较低。
