Towards a Question/Answering System in Moroccan Legal Domain: data preparation and question classification phase using ML approaches

This paper is a part of larger work aiming the construction of a Question/Answering System with Moroccan legal domain. It concerns mainly the phase of data preparation and question classification. The classification of legal questions is a crucial step in a legal question-answering system. It can help to find the relevant documents that contain the answers. With the increasing volume of legal data, legal question classification can be beneficial for legal practitioners to efficiently identify relevant laws, precedents, and regulations applied to a specific case. Machine learning algorithms and Natural Language Processing (NLP) techniques can effectively process large amounts of unstructured legal data, which can be difficult to be processed manually. In this work, we investigate the use of different word representation techniques and different stemmers in the preprocessing stage, aiming to classify legal questions using three machine learning algorithms, Support Vector Machine, Naïve Bayes, and Random Forest. The experimental results show that the Random Forest algorithm achieved the highest accuracy and precision. Additionally, they confirm the significant impact of question preprocessing on the classification results enhancement.