Prediction of court decision from Arabic documents using deep learning

Abstract

AbstractThe increasing amount of electronic legal documents represents a great opportunity for the development of intelligent computational systems for legal texts processing and classification. Most of these systems use classical machine learning and large datasets in English. This paper proposes an approach to automatically predict legal case outcome from written description of the events in Arabic using deep learning. An in‐house corpus from the decisions of the Moroccan Court of Cassation is built and used to train a deep learning model that predicts judgement. As the created corpus is of limited size, a new data augmentation method is proposed to boost the prediction performance. Two settings for text representation are tested, namely FastText and GloVe embeddings, and multiple deep learning models architectures are tested. The proposed approach succeeds in predicting judicial decisions of the Moroccan Court of Cassation with an accuracy of 80.51% on six classes. Even with a small dataset, the proposed data augmentation method was helpful in improving the overall models' performance. Despite the advancement in the area of legal judgement prediction over the years, this work is the first attempt to predict legal outcome using the documents of the Moroccan Cassation court. The corpus created in the context of this work will be made publicly available to the community.