Abstract

This paper studies the challenges and progressions of Legal Judgment Prediction, in particular on the grounds of improving efficiency, accuracy, and fairness in judicial systems. The study examines five sub-research questions: the limitation of existing LJP methods, the role of factual logic in judgment reasoning, the integration of external legal knowledge, the effectiveness of a chain prompt reasoning module, and the impact of contrastive knowledge fusion on long-tail cases. A qualitative research methodology is followed to design and validate an interpretable framework for LJP, featuring a chain prompt reasoning module to strengthen factual logic and a contrastive knowledge fusing module to incorporate external legal knowledge. Results indicate notable improvements in the prediction accuracy, interpretability, and handling of complex long-tail cases. Despite the specific datasets used, the proposed framework is a demonstration of its potential in wider applications and theoretical contributions to legal AI. Future work will be based on diverse data sources and methodologies to generalize and improve these findings.