Justice AI: Legal Case Retrieval Using Dense Passage Retrieval

초록

In contemporary society, resolving legal issues effectively requires precise knowledge and expert assistance, yet accessing and utilizing legal information remains challenging for many. Justice AI addresses this gap using advanced Dense Passage Retrieval (DPR) technology, implemented with the BERT-based KoBERT model and the GPT-based LCube model. KoBERT, optimized for Korean legal texts, ensures high relevance and comprehension of complex legal terms, while LCube enhances understanding and generates natural language responses for deeper contextual insight. The system achieved a cosine similarity score of 0.9002, demonstrating excellent relevance in retrieved results. Performance evaluation using Precision, Recall, and F1 Score yielded metrics of 0.42, 1.0, and 0.5915, respectively, for LCube. Although tailored for Korean legal texts, the underlying technology showcases strong adaptability for diverse legal systems, positioning Justice AI as a globally applicable tool with potential to transform legal research and services. © 2025 IEEE.

키워드