RLJP: Legal Judgment Prediction via First-Order Logic Rule-enhanced with Large Language Models

TL;DR

The paper tackles Legal Judgment Prediction (LJP) by addressing the missing adaptive reasoning in prior models. It introduces RLJP, a three-stage framework that initializes First-Order Logic (FOL) judgment rules with an LLM, optimizes them through Confusion-Aware Contrastive Learning (CACL) using confusable cases organized via a tree-splitting strategy, and finally applies the refined rules in an examination stage with a lightweight label scorer. Empirical evaluations on two Chinese LJP datasets (CAIL2018 and CJO22) show state-of-the-art performance across all metrics, with especially large gains on long, complex cases, validating the utility of dynamic, logic-guided reasoning in legal prediction. The approach advances practical LJP by enabling case-specific interpretability and adaptive reasoning, while code availability supports reproducibility and future extension in multi-language or multi-task settings. The findings underscore the potential of integrating symbolic legal reasoning with large-language models to improve accuracy and robustness in legally critical predictions.

Abstract

Legal Judgment Prediction (LJP) is a pivotal task in legal AI. Existing semantic-enhanced LJP models integrate judicial precedents and legal knowledge for high performance. But they neglect legal reasoning logic, a critical component of legal judgments requiring rigorous logical analysis. Although some approaches utilize legal reasoning logic for high-quality predictions, their logic rigidity hinders adaptation to case-specific logical frameworks, particularly in complex cases that are lengthy and detailed. This paper proposes a rule-enhanced legal judgment prediction framework based on first-order logic (FOL) formalism and comparative learning (CL) to develop an adaptive adjustment mechanism for legal judgment logic and further enhance performance in LJP. Inspired by the process of human exam preparation, our method follows a three-stage approach: first, we initialize judgment rules using the FOL formalism to capture complex reasoning logic accurately; next, we propose a Confusion-aware Contrastive Learning (CACL) to dynamically optimize the judgment rules through a quiz consisting of confusable cases; finally, we utilize the optimized judgment rules to predict legal judgments. Experimental results on two public datasets show superior performance across all metrics. The code is publicly available{https://anonymous.4open.science/r/RLJP-FDF1}.

Figures (6)

• Figure 1: RLJP framework. The blue box is the Rule Initialization Module on the upper left(§\ref{['sec:3.2']}), the gray box is the Rule Optimization Module(§\ref{['sec:3.3']}) with Confusion-Aware Contrastive Learning, and the green box is the Examination Module(§\ref{['sec:3.4']}) for the completion of LJP tasks based on the optimized rules.
• Figure 2: The construction of quiz experience.
• Figure 3: The prompt template for optimization direction generation.
• Figure 4: The prompt template for the quiz in the confusable case set.
• Figure 5: The prompt template for optimization direction generation.