🤖 AI Summary
To address the ranking performance bottleneck in similar-case retrieval for legal AI, this paper proposes a novel learning-to-rank paradigm that bypasses the final classification layer of language models—thereby mitigating overfitting induced by class imbalance in fine-tuning. Specifically, RankSVM is introduced into the Chinese legal domain to replace the standard fully connected layer in traditional fine-tuning pipelines. We present the first systematic pairwise ranking framework integrating BERT/ERNIE with RankSVM, and rigorously evaluate it on the LeCaRDv1 and LeCaRDv2 benchmarks. Experimental results demonstrate consistent and statistically significant improvements: average gains of 2.3%–4.1% in both NDCG@5 and MAP, alongside enhanced model generalization. The implementation is publicly available.
📝 Abstract
Given the rapid development of Legal AI, a lot of attention has been paid to one of the most important legal AI tasks--similar case retrieval, especially with language models to use. In our paper, however, we try to improve the ranking performance of current models from the perspective of learning to rank instead of language models. Specifically, we conduct experiments using a pairwise method--RankSVM as the classifier to substitute a fully connected layer, combined with commonly used language models on similar case retrieval datasets LeCaRDv1 and LeCaRDv2. We finally come to the conclusion that RankSVM could generally help improve the retrieval performance on the LeCaRDv1 and LeCaRDv2 datasets compared with original classifiers by optimizing the precise ranking. It could also help mitigate overfitting owing to class imbalance. Our code is available in https://github.com/liuyuqi123study/RankSVM_for_SLR