CaseEncoder: A Knowledge-enhanced Pre-trained Model for Legal Case Encoding

TL;DR

The paper addresses the challenge that general-domain PLMs inadequately capture legal reasoning for case retrieval. It proposes CaseEncoder, a knowledge-enhanced encoder that uses fine-grained sampling of unambiguous law articles and a MLM plus Biased Circle Loss pre-training objective to learn embeddings aligned with legal relevance. The method achieves significant zero-shot gains on LeCaRD and CAIL benchmarks and shows embedding quality improvements for downstream tasks like charge prediction. This work highlights the practical value of injecting domain-specific legal knowledge into both data construction and pre-training for legal NLP.

Abstract

Legal case retrieval is a critical process for modern legal information systems. While recent studies have utilized pre-trained language models (PLMs) based on the general domain self-supervised pre-training paradigm to build models for legal case retrieval, there are limitations in using general domain PLMs as backbones. Specifically, these models may not fully capture the underlying legal features in legal case documents. To address this issue, we propose CaseEncoder, a legal document encoder that leverages fine-grained legal knowledge in both the data sampling and pre-training phases. In the data sampling phase, we enhance the quality of the training data by utilizing fine-grained law article information to guide the selection of positive and negative examples. In the pre-training phase, we design legal-specific pre-training tasks that align with the judging criteria of relevant legal cases. Based on these tasks, we introduce an innovative loss function called Biased Circle Loss to enhance the model's ability to recognize case relevance in fine grains. Experimental results on multiple benchmarks demonstrate that CaseEncoder significantly outperforms both existing general pre-training models and legal-specific pre-training models in zero-shot legal case retrieval.

Figures (3)

• Figure 1: The overall framework of CaseEncoder.
• Figure 2: An illustration of the process of collecting a fine-grained unambiguous article corpus.
• Figure 3: The visualization of case embeddings generated by four PLMs in the zero-shot manner.