Classification of Civil Court Decision Documents Using Legal BERT and IndoLegal-BERT

Classifying civil court decisions at the first level is difficult, especially when distinguishing between types of decisions and court orders. This law is significant because civil cases considerably impact individuals, organizations, and society. This research uses $\mathbf{5 0 0}$ court decisions and judge rulings for civil instances taken from the Supreme Court of the Republic of Indonesia’s website. The models used in the classification process are LEGAL-BERT and IndoLEGAL-BERT. The research employs two model evaluation scenarios, before and after fine-tuning, to assess each model’s precision, recall, and accuracy performance. The study focuses on classification based on selected features, namely the verdict and the core of the claim. The evaluation results indicate that the IndoLEGAL-BERT model performs better than the LEGAL-BERT model in all conditions. The reason is that IndoLEGAL-BERT is a pre-trained model that has used Indonesian language documents and is specific to the legal domain.