BiCA: Effective Biomedical Dense Retrieval with Citation-Aware Hard Negatives
Abstract
BiCA uses citation links to improve biomedical retrieval models by providing effective hard negatives, enhancing zero-shot and long-tailed performance with minimal fine-tuning.
Hard negatives are essential for training effective retrieval models. Hard-negative mining typically relies on ranking documents using cross-encoders or static embedding models based on similarity metrics such as cosine distance. Hard negative mining becomes challenging for biomedical and scientific domains due to the difficulty in distinguishing between source and hard negative documents. However, referenced documents naturally share contextual relevance with the source document but are not duplicates, making them well-suited as hard negatives. In this work, we propose BiCA: Biomedical Dense Retrieval with Citation-Aware Hard Negatives, an approach for hard-negative mining by utilizing citation links in 20,000 PubMed articles for improving a domain-specific small dense retriever. We fine-tune the GTE_small and GTE_Base models using these citation-informed negatives and observe consistent improvements in zero-shot dense retrieval using nDCG@10 for both in-domain and out-of-domain tasks on BEIR and outperform baselines on long-tailed topics in LoTTE using Success@5. Our findings highlight the potential of leveraging document link structure to generate highly informative negatives, enabling state-of-the-art performance with minimal fine-tuning and demonstrating a path towards highly data-efficient domain adaptation.
Community
This paper presents a new way to mine hard-negatives, using citation networks and graph traversal methods. Fine-tuning dense retrievers leads to strong performance gains on BEIR and LoTTE. We also publish our model and datasets ☺️ .
This is an automated message from the Librarian Bot. I found the following papers similar to this paper.
The following papers were recommended by the Semantic Scholar API
- ModernBERT + ColBERT: Enhancing biomedical RAG through an advanced re-ranking retriever (2025)
- Enhancing Transformer-Based Rerankers with Synthetic Data and LLM-Based Supervision (2025)
- E2Rank: Your Text Embedding can Also be an Effective and Efficient Listwise Reranker (2025)
- PairSem: LLM-Guided Pairwise Semantic Matching for Scientific Document Retrieval (2025)
- RAG-BioQA Retrieval-Augmented Generation for Long-Form Biomedical Question Answering (2025)
- CustomIR: Unsupervised Fine-Tuning of Dense Embeddings for Known Document Corpora (2025)
- Doc2Query++: Topic-Coverage based Document Expansion and its Application to Dense Retrieval via Dual-Index Fusion (2025)
Please give a thumbs up to this comment if you found it helpful!
If you want recommendations for any Paper on Hugging Face checkout this Space
You can directly ask Librarian Bot for paper recommendations by tagging it in a comment:
@librarian-bot
recommend
Models citing this paper 2
Datasets citing this paper 2
Spaces citing this paper 0
No Space linking this paper
Collections including this paper 0
No Collection including this paper