Legal Document Information Retrieval using Long Short Term Memory-Based Domain Classification Technique
Rasmi Rani Dhala, A. V. S. Pavan Kumar, Soumya Priyadarsini Panda · 2025
This work presents a legal document information retrieval system that retrieves the most relevant documents quickly from a collection of legal reports and documents. For this purpose, a legal document repository is created by collecting the documents and case study reports of different legal matters of last five years. To minimize the search space and time in retrieving the required documents, a long short-term memory (LSTM) based domain classification method is used that determines the relevance of the legal queries to appropriate legal sub- domains. The retrieval model was tested in several context to evaluate the performance of the legal model. The domain classification approach is benchmarked against a number of classifiers (K-nearest neighbors, logistic regression, random forest and XGBoost) trained in exactly the same settings. The model achieves an average accuracy of 97.82%, precision of 97.73% and recall of 97. 88%. The overall performance of the legal retrieval model is also compared with the Boolean retrieval model for the considered legal sub domains. The model for the retrieval of legal document as described obtains a relative accuracy gain of approximately 11% compared to the traditional approach, demonstrating the efficiency of the proposed method on the legal sub-domains.