<![CDATA[This study explored the application of the GPT API framework to automate case complexity classification and topic modelling in Indonesian Tax Court decisions. Using a dataset of 5,000 anonymized tax dispute summaries, we designed a prompt-based classification pipeline supported by expert-labelled benchmarks. The case complexity classification task, categorizing cases into low and high complexity. The GPT-based model achieving 87% precision. This indicates the model’s practical ability to simulate legal judgment in triaging case difficulty. Simultaneously, topic modelling was performed to identify key dispute themes across grouped cases. The three most frequently recurring themes were: (1) input VAT correction errors in VAT disputes (26.1%), (2) net income adjustments in income tax cases (9.1%), and (3) customs valuation issues in import transactions (17.4%). These model-derived clusters aligned closely with expert taxonomies and provided useful summaries of dispute patterns over time. The methodology built using Python, Google Colab, and the OpenAI GPT API. By structuring Indonesia’s growing corpus of tax litigation into actionable categories, this approach strengthens the country’s digital justice transformation. It enables better resource allocation and faster dispute resolution.]]>
Copyrights © 2025
