Enhancing Judgment Document Generation via Agentic Legal Information Collection and Rubric-Guided Optimization

2026-05-03

AI Summary

This work addresses key challenges in automatic legal judgment generation, including incomplete legal retrieval, hallucinated statutory citations, and logical inconsistencies. To overcome these issues, the authors propose Judge-R1, a novel framework that integrates agent-driven multi-source legal information retrieval with an optimization mechanism grounded in judicial evaluation criteria. The retrieval component employs a dynamically planned agent to achieve precise identification of relevant legal provisions, while the generation process is refined through Group Relative Policy Optimization guided by a legally informed reward function that enhances logical coherence and regulatory compliance. Evaluated on the JuDGE benchmark, Judge-R1 significantly outperforms existing methods in both legal accuracy and overall generation quality, establishing a new state-of-the-art performance.

Abstract

Automating the drafting of judgment documents is pivotal to judicial efficiency, yet it remains challenging due to the dual requirements of comprehensive retrieval of legal information and rigorous logical reasoning. Existing approaches, typically relying on standard Retrieval-Augmented Generation and Supervised Fine-Tuning, often suffer from insufficient evidence recall, hallucinated statutory references, and logically flawed legal reasoning. To bridge this gap, we propose Judge-R1, a unified framework designed to enhance LLM-based judgment document generation by jointly improving legal information collection and judgment document generation. First, we introduce Agentic Legal Information Collection, which employs a dynamic planning agent to retrieve precise statutes and precedents from multiple sources. Second, we implement Rubric-Guided Optimization, a reinforcement learning phase utilizing Group Relative Policy Optimization (GRPO) with a comprehensive legal reward function to enforce adherence to judicial standards and reasoning logic. Extensive experiments on the JuDGE benchmark demonstrate that Judge-R1 significantly outperforms state-of-the-art baselines in both legal accuracy and generation quality.

Problem

• judgment document generation
• legal information retrieval
• legal reasoning
• statutory reference hallucination
• judicial efficiency

Innovation

• Agentic Legal Information Collection
• Rubric-Guided Optimization
• Retrieval-Augmented Generation
• Group Relative Policy Optimization
• Legal Reasoning