Guang Yang, Xin Zhang, Feng Yao
Legal case retrieval (LCR) is a fundamental task in the field of legal intelligence research and holds a foundational position in both academic research and practical work in the legal domain. The academic community has carried out relevant research on this task at a very early stage. In particular, after the emergence of neural networks (NNs), academic achievements have continuously emerged, and relevant researchers have already published surveys related to LCR. However, existing surveys still have several problems, including unclear task definition, incomplete coverage of approaches, and failure to keep pace with the academic frontier. To address these problems, our survey focuses on NN-based methods and provides a narrative review of LCR regarding the aspects of task definition, methods, datasets and evaluation metrics, typical applications, and future research directions. This survey makes the following contributions: (1) a complete task definition for LCR; (2) a comprehensive summary of NN-based methods, especially focusing on the achievements of Large Language Models (LLMs); (3) comprehensive and clear organization of existing datasets and evaluation metrics; and (4) reasonable analysis of typical applications and future prospects.