Legal Judgment Prediction Incorporating Guiding Cases Matching

Hengzhi Li, Shubin Cai, Zhong Ming

Abstract: Legal judgment prediction aims to predict the judgment result based on the case fact description. It is an important application of natural language processing within the legal field. To enhance the impartiality and consistency of the judiciary, the Supreme People’s Court selects representative cases and categorizes them as guiding or typical cases. However, the current research is based on the unified training of large-scale datasets, and does not distinguish the value of different existing cases. To address these issues, this paper collects relevant guiding cases and typical cases, designs a legal judgment prediction method that incorporates matching guiding cases, and proposes a legal judgment prediction model called MAGIC (MAtching GuIding Cases), based on multi-feature fusion. The goal of this research is to improve the prediction performance of legal judgments by utilizing the value of high-quality cases. The experimental results on the CAIL2018 evaluation datasets demonstrate that the proposed method significantly improves the performance of legal judgment prediction models.