跳转至

CoAL-RAG：一种兼顾复杂度感知的法律检索增强生成方法

文章背景与核心概要

法律咨询问题往往呈现出不同程度的复杂性。传统的检索增强生成（RAG）系统通常采用单一、统一的检索策略，这往往会导致简单查询的“过度推理”以及复杂案件的“解释性不足”，难以满足高风险法律领域对回答质量和系统效率的双重需求。

为了克服这些挑战，研究人员推出了 CoAL-RAG（Complexity-Aware Legal Retrieval-Augmented Generation，复杂度感知法律检索增强生成）。该方法的核心是一个由两大支柱驱动的多维度评估机制：1. 问题本质：根据输入查询的逻辑结构量化精确的推理需求；2. 检索一致性：衡量语义检索与关键词检索之间的差异，以隐式反映问题的复杂度。

通过这种评估，CoAL-RAG 能够自适应地路由检索策略并动态过滤上下文信息。实验结果表明，CoAL-RAG 在中文法律基准（SocialLawQA、LawBench）上的表现显著优于基准模型，并在英文数据集（LexGLUE、CaseHold）上展现出卓越的跨司法管辖区泛化能力。

CoAL-RAG: A Complexity-Aware Legal Retrieval-Augmented Generation Method

Summary

Legal consultation questions often present varying degrees of complexity. Traditional Retrieval-Augmented Generation (RAG) systems typically rely on a single, uniform retrieval strategy, which often results in over-reasoning for simple inquiries and insufficient interpretability for complex cases. Consequently, they struggle to meet the dual demands of high answer quality and system efficiency in high-risk legal domains.

To overcome these challenges, researchers introduce CoAL-RAG (Complexity-Aware Legal Retrieval-Augmented Generation). This method features a multi-dimensional evaluation mechanism driven by two primary pillars: 1. Question Essence: Quantifies the precise reasoning demands based on the logical structure of the input query. 2. Retrieval Consistency: Measures the discrepancy between semantic retrieval and keyword retrieval to implicitly reflect problem complexity.

Through this evaluation, CoAL-RAG adaptively routes retrieval strategies and dynamically filters contextual information. Experimental results show that CoAL-RAG outperforms baseline models significantly on Chinese legal benchmarks (SocialLawQA, LawBench) and exhibits exceptional cross-jurisdictional generalization on English datasets (LexGLUE, CaseHold).

Paper Metadata

Paper Metadata

  • Subject Areas: Computation and Language (cs.CL), Artificial Intelligence (cs.AI)
  • Subject Areas: Computation and Language (cs.CL), Artificial Intelligence (cs.AI)
  • Authors:
  • Jin Su
  • Zhuofeng Zhao
  • Huanhuan Wang
  • Hao Chen
  • Authors:
  • Jin Su
  • Zhuofeng Zhao
  • Huanhuan Wang
  • Hao Chen
  • Submission Date: August 18, 2026
  • Submission Date: August 18, 2026

Abstract

Abstract

Legal consultation questions exhibit multi-level complexity. A single retrieval strategy often leads to over-reasoning for simple questions and poor interpretability for complex ones, making it difficult to meet the requirements for both answer quality and efficiency in high-risk scenarios. To address this issue, this paper proposes CoAL-RAG, a complexity-aware legal retrieval-augmented generation method, which constructs a multi-dimensional evaluation mechanism based on "question essence" and "retrieval consistency" to enable adaptive routing of retrieval strategies. First, the reasoning demand is quantified according to the logical structure of the question. Then, the discrepancy between semantic retrieval and keyword retrieval is utilized to indirectly reflect problem complexity, thereby selecting the most appropriate retrieval strategy and dynamically filtering contextual information. Experimental results demonstrate that the proposed method significantly outperforms baseline models not only on Chinese legal benchmarks (SocialLawQA, LawBench) but also demonstrates strong cross-jurisdictional generalization on English datasets (LexGLUE, CaseHold). Specifically, on Chinese datasets, the BLEU score improves by 42.5\% and ROUGE-L reaches 3.6 times that of knowledge graph-based methods. On English benchmarks, CoAL-RAG maintains highly competitive accuracy, achieving an optimal balance between generation quality, deep logical reasoning, and system efficiency across different legal systems.

Key Performance Highlights

Key Performance Highlights

  • Chinese Legal Benchmarks (SocialLawQA, LawBench):
  • BLEU Score: Improved by 42.5% over baseline models.
  • ROUGE-L: Reached 3.6 times the performance of traditional knowledge graph-based approaches.
  • Chinese Legal Benchmarks (SocialLawQA, LawBench):
  • BLEU Score: Improved by 42.5% over baseline models.
  • ROUGE-L: Reached 3.6 times the performance of traditional knowledge graph-based approaches.
  • English Legal Benchmarks (LexGLUE, CaseHold):
  • Maintained highly competitive accuracy, verifying strong cross-jurisdictional generalization across disparate legal systems.
  • English Legal Benchmarks (LexGLUE, CaseHold):
  • Maintained highly competitive accuracy, verifying strong cross-jurisdictional generalization across disparate legal systems.
  • System Balance: Successfully established an optimal equilibrium connecting generation quality, deep logical reasoning, and processing efficiency.
  • System Balance: Successfully established an optimal equilibrium connecting generation quality, deep logical reasoning, and processing efficiency.

Links & Resources