Optimizing Legal Document Retrieval in Vietnamese with Semi-Hard Negative Mining

2025-07-19

AI Summary

To address the domain-knowledge deficiency and low retrieval accuracy of large language models in Vietnamese legal document retrieval, this paper proposes a two-stage retrieval framework: (1) an efficient candidate retrieval stage using a fine-tuned Bi-Encoder, and (2) a fine-grained re-ranking stage employing a Cross-Encoder. Key innovations include semi-hard negative mining and fine-grained negative sampling, a novel Exist@m evaluation metric, and a customized loss function to mitigate training bias and enhance robustness. The lightweight, single-pass architecture achieves top-three performance in the SoICT Hackathon 2024 legal retrieval task—matching the accuracy of complex ensemble models while reducing parameter count significantly. This demonstrates the method’s effectiveness and practicality for specialized, low-resource language legal domains.

Technology Category

• Data Mining & Knowledge Management: Conversational Systems for Recommendation & Retrieval
• Computer Vision: Image and Video Retrieval
• Natural Language Processing: Information Extraction

Application Category

• Search and Retrieval-Augmented AI: Vertical and domain-specific search
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Large Language Models (LLMs) face significant challenges in specialized domains like law, where precision and domain-specific knowledge are critical. This paper presents a streamlined two-stage framework consisting of Retrieval and Re-ranking to enhance legal document retrieval efficiency and accuracy. Our approach employs a fine-tuned Bi-Encoder for rapid candidate retrieval, followed by a Cross-Encoder for precise re-ranking, both optimized through strategic negative example mining. Key innovations include the introduction of the Exist@m metric to evaluate retrieval effectiveness and the use of semi-hard negatives to mitigate training bias, which significantly improved re-ranking performance. Evaluated on the SoICT Hackathon 2024 for Legal Document Retrieval, our team, 4Huiter, achieved a top-three position. While top-performing teams employed ensemble models and iterative self-training on large bge-m3 architectures, our lightweight, single-pass approach offered a competitive alternative with far fewer parameters. The framework demonstrates that optimized data processing, tailored loss functions, and balanced negative sampling are pivotal for building robust retrieval-augmented systems in legal contexts.

Problem

• Enhancing legal document retrieval efficiency and accuracy
• Mitigating training bias with semi-hard negative mining
• Optimizing lightweight models for legal domain precision

Innovation

• Fine-tuned Bi-Encoder for rapid retrieval
• Cross-Encoder for precise re-ranking
• Semi-hard negative mining to reduce bias