Reproduce “Yes, BM25 is a Strong Baseline for Legal Case Retrieval” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Guilherme Moraes Rosa, Ruan Chaves Rodrigues, Roberto Lotufo, Rodrigo Nogueira
Researchers demonstrated that a carefully adapted BM25 algorithm achieved the second-best F1 score in the COLIEE 2021 legal case retrieval task. This work highlights the strong performance of a classic, unsupervised model in a specialized, complex domain.
Key insights
- BM25 remains a highly competitive baseline for legal case retrieval, effectively challenging the assumption that only complex neural models can achieve top performance.
- Clever adaptation, such as document segmentation and max-score aggregation, can significantly enhance traditional IR models' effectiveness on long, complex texts.
- Computational constraints sometimes necessitate pragmatic trade-offs, like truncating input segments, without entirely sacrificing competitive performance.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.