Research map / Label-efficient and robust learning / Domain adaptation and generalization
BiCA: Effective Biomedical Dense Retrieval with Citation-Aware Hard Negatives
Aarush Sinha, Pavan Kumar S, Roshan Balaji, Nirav Pravinbhai Bhatt
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| fine-tune GTE-small on citation-aware hard negatives | GTE-small · NQ | nDCG@10 (x100): 32.0 -> 50.2 | Improved |
| fine-tune GTE-base on citation-aware hard negatives | GTE-base · ArguAna | nDCG@10 (x100): 41.0 -> 57.1 | Improved |
| fine-tune GTE-small on citation-aware hard negatives | GTE-small · SciFact | nDCG@10 (x100): 72.7 -> 72.7 | No clear effect |
| fine-tune GTE-small on citation-aware hard negatives | GTE-small · NFCorpus | nDCG@10 (x100): 34.9 -> 34.7 | No clear effect |
| fine-tune on citation-aware negatives (average over 14 BEIR tasks) | GTE-base · BEIR 14 tasks | nDCG@10 gain: +6.8 | Improved |
| use 3 traversal paths instead of 2 (ablation) | BiCA mining, bert-base · 5 BEIR tasks avg | nDCG@10: 0.2068 -> 0.2739 | Improved |
| use 3 traversal paths instead of 4 (ablation) | BiCA mining, bert-base · 5 BEIR tasks avg | nDCG@10: 0.2184 -> 0.2739 | Improved |
| use 3 traversal paths instead of 5 (ablation) | BiCA mining, bert-base · 5 BEIR tasks avg | nDCG@10: 0.2734 -> 0.2739 | No clear effect |
| use path length 3 instead of 2 (ablation) | BiCA mining, bert-base · 5 BEIR tasks avg | nDCG@10: 0.2041 -> 0.2739 | Improved |
| use path length 3 instead of 5 (ablation) | BiCA mining, bert-base · 5 BEIR tasks avg | nDCG@10: 0.2580 -> 0.2739 | Improved |
| train on 20k triplets instead of 1k (ablation) | bert-base-uncased · ArguAna | nDCG@10: 0.384 -> 0.444 | Improved |
| train on 20k triplets instead of 1k (ablation) | bert-base-uncased · SciFact | nDCG@10: 0.262 -> 0.493 | Improved |
Closest papers by meaning
- TriSampler: A Better Negative Sampling Principle for Dense Retrieval (AAAI 2024 · similarity 0.84)
- HLM-Cite: Hybrid Language Model Workflow for Text-based Scientific Citation Prediction (NeurIPS 2024 · similarity 0.83)
- Reasoning-Enhanced Healthcare Predictions with Knowledge Graph Community Retrieval (ICLR 2025 · similarity 0.83)
- Mitigating the Impact of False Negative in Dense Retrieval with Contrastive Confidence Regularization (AAAI 2024 · similarity 0.82)
- Combining Multiple Supervision for Robust Zero-Shot Dense Retrieval (AAAI 2024 · similarity 0.82)
- Scalable In-context Ranking with Generative Models (NeurIPS 2025 · similarity 0.81)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
