Budget Analysis in the Social Policy Field Using Text Data and Fiscal Information

This study analyzed government detailed project budgets by combining AI, big data, and expert judgments. Instead of traditional classifications, the study used the 27 social policy agendas from the announced '2023 Core Social Policy Implementation Plan' to categorize government projects. Additionally, the life cycle was used as a classification criterion. Natural language processing(NLP) technology was employed to understand and classify textual data describing detailed projects, successfully classifying government projects and budgets from 2020 to 2023 according to the 27 agendas. Public data from 'NKIS' and 'Open Finance' were utilized in the classification, and KeyBERT was used for NLP. The classification results allowed the identification of annual changes in the number and budget of government projects according to the 27 agendas, as well as the degree of imbalance in detailed projects for each agenda. Furthermore, the classification results by life cycle provided insights into who the detailed projects and budgets are intended for. While NLP played a key role in the results, expert knowledge and judgment were crucial. The research findings suggest evidence for making judgments on efficient budget execution and interagency cooperation. The study also hints at the potential for more in-depth, field-specific research on the 27 social policy issues and life cycle.