Reproduce “BiCA: Effective Biomedical Dense Retrieval with Citation-Aware Hard Negatives” 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
Aarush Sinha, Pavan Kumar S, Roshan Balaji, Nirav Pravinbhai Bhatt
BiCA introduces a novel strategy for generating hard negatives in dense retrieval by leveraging multi-hop citation chains from biomedical literature, leading to state-of-the-art zero-shot performance on diverse retrieval benchmarks and improved computational efficiency.
Key insights
- Multi-hop citation links inherently provide semantically challenging yet distinct documents that serve as highly effective hard negatives for dense retrieval model training.
- Training with these citation-aware hard negatives enables models to achieve state-of-the-art retrieval performance with significantly fewer parameters compared to many larger baselines.
- The fine-tuning approach, though initialized with biomedical data, leads to models with robust generalizability, performing strongly across both domain-specific and general-domain retrieval tasks.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.