A Legal Consultant Chatbot for Supreme Court of Oromia Using Deep Learning Techniques

Abstract

Legal awareness is crucial in promoting the rule of law and fostering legal cultural awareness. The legal domain is characterized by complex and intricate language structures, making it challenging for non-experts to navigate. The introduction of generative chatbots for court services has the potential to revolutionize the way legal information is accessed and understood by the general public. Many court clients miss their right due to lack of awareness about legal. In this study, we present the development of a generative legal consultant chatbot for the Oromia Supreme Court, aimed at providing accurate, timely, and accessible legal information to users. We collected 2MB of data from primary and secondary (Facebook, Telegram, and websites) sources, to train our models. The collected dataset was prepared in the form of plaintext and preprocessed using various NLP techniques. An experimental approach was performed in this study to determine best model. Deep learning models based on GPT (Generative Pre-trained Transformer) and LSTM (Long Short-Term Memory) with ELMO (Embedding’s from Language Models Representation), have been experimented for legal consultant services using 10-fold stratified cross-validation on fine-tuned hyperparameters. Different classification metrics are used for evaluating models. The GPT model outperformed the LSTM with ELMO model, achieving a remarkable accuracy of 90.8%. Finally, the best performed model was evaluated by legal experts and achieved 88.5% accuracy. Eventually, our work revealed that using language model with fine-tuned hyperparameters significantly enhances the quality of legal consultant chatbot services. The developed legal consultant chatbot demonstrates its potential to support the Oromia Supreme Court in providing efficient and accessible legal services. Future work may involve expanding the dataset, fine-tuning the models, and integrating additional features to further enhance the chatbot's performance and accuracy.