Adaptive Multi-Strategy Legal Case Retrieval and Outcome Prediction: A Story-Based Approach for Indian Courts

The prediction of legal case outcomes and the retrieval of precedents are identified as central challenges in the field of computational law, particularly in jurisdictions with a strong case law tradition, such as India. Conventional methods are observed to rely primarily on semantic similarity or citation networks, at the expense of narrative structure and factual context, which are utilised by legal practitioners in the discovery of relevant precedents. An adaptive multi-strategy framework is presented in this paper, wherein seven complementary matching strategies are combined, including semantic understanding using Legal-BERT (Bidirectional Encoder Representations from Transformers) embeddings, legal terminology matching using Term Frequency Inverse Document Frequency (TF-IDF) vectorization, story narrative analysis, extraction of factual elements, intelligent comparison of financial amounts, location-based matching, and analysis of legal metadata using case-type-adaptive weighting. The framework was designed to provide retrieval of the top-15 most similar cases ranked according to a global multi-dimensional similarity measure, as well as ensemble-based prediction of outcomes with detailed explanations. The assessment was conducted on actual Indian legal cases of varying types, including motor accident, compensation, criminal appeals, civil disputes, and property matters. High effectiveness was demonstrated in narrative-based precedent matching, and the relative importance of each strategy was observed to be dynamically adjusted by the adaptive weighting mechanism based on the detected case type. Key gaps in computational law are addressed by the framework, as precedents are discovered on the basis of factual similarity rather than statutory references alone, thereby facilitating more informed legal research and decision-making.