Summarization, Prediction, and Analysis of Turkish Constitutional Court Decisions With Explainable Artificial Intelligence and a Hybrid Natural Language Processing Method

The use of artificial intelligence in legal analysis represents a significant transformation process in modern legal practices. The literature review showed that the number of studies on Turkish court decisions was limited, and only classification applications were developed. This study aims to summarize, predict, and analyze the Turkish Constitutional Court’s decision texts. In this context, first a new, unique dataset was created. The dataset included class labels, original court decision texts, and summary court decision texts created by expert lawyers. Then, a hybrid summary model was established, combining extractive and abstract summarization to summarize long court decisions automatically. With this model, the summarization efficiency was increased, and the token limitation problem common in existing transformative models was eliminated. The model showed good performance for summarization by obtaining Rouge-1, Rouge-2, and Rouge-L scores of 0.6129, 0.5884, and 0.5891 respectively. After the summarization phase, the Legal Judgment Prediction applications were developed. Separate classification models were developed for court decision texts and summary court decision texts, obtained using the hybrid model. When the models were compared, the XGBoost model achieved the best performance in legal judgment prediction tasks, with an accuracy rate of 93.84% for full texts and 62.30% for summary texts. In the final stage of the study, the model results were explained using the SHapley Additive exPlanations method. The findings of this study emphasize the superiority of hybrid approaches to legal document analysis. They highlighted the vital role of explainable techniques in improving transparency and reliability in legal processes.