Legal-LM: Knowledge Graph Enhanced Large Language Models for Law Consulting

Juanming Shi, Qinglang Guo, Yong Liao, Yuxing Wang, Shijia Chen, Shenglin Liang

Abstract: This paper introduces Legal-LM, an advanced Large Language Model (LLM) enhanced with a Knowledge Graph, specifically designed for legal consulting in the Chinese legal domain. Addressing the challenges of domain-specific adaptation, data veracity, and consultations with non-professional users in legal AI, Legal-LM incorporates extensive legal corpora and a knowledge graph for effective legal knowledge acquisition. The model utilizes techniques such as external legal knowledge basis, soft prompts, and Direct Preference Optimization (DPO) to ensure accurate and diverse legal advice. Our experimental results demonstrate that Legal-LM exhibits superior performance over existing models in legal question answering, case analysis, and legal recommendations, these show its potential to facilitate legal consulting and education.