TaSbeeb: A judicial decision support system based on deep learning framework

Abstract

Since the early 1980s, the legal domain has shown a growing interest in Artificial Intelligence approaches to tackle the increasing number of cases worldwide. TaSbeeb is a deep learning (DL)-based judicial decision support system (JDSS) designed for legal professionals in Saudi courts by retrieving judicial reasoning, Qur’anic verses, and hadiths from a knowledge base. The proposed system consists of three phases: annotation, classification, and information retrieval. To annotate judicial text, we developed Ann-Judicial, a semi-automatic method. To handle the imbalanced corpus for classification, we devised homogeneous and heterogeneous stacking DL models. For information retrieval, we proposed Jud_RoBERTa, a judicial language model. TaSbeeb achieved high accuracy and F-scores in both the classification and information retrieval blocks, showing good accuracy despite complexities in the judicial field and interference between cases. Specifically, the classification phase achieved an accuracy and F-score of 95.8%, while the information retrieval phase achieved an accuracy of 79.8% and F-score of 79.3%. The proposed JDSS has potential for extension to other courts and can be used in judicial inspection. TaSbeeb represents a significant stride towards a more efficient and accurate judicial decision-making process in the Arabic legal system, which has been hindered by a lack of research on Arabic JDSS.