New REPAIR framework boosts scientific retrieval accuracy · 2 sources tracked

Researchers have developed REPAIR, a novel data augmentation framework designed to improve the accuracy of scientific information retrieval systems. This self-evolving framework addresses challenges posed by long-tailed concepts and the fact-sensitive nature of scientific corpora, which often hinder dense retrievers and LLM integrations. REPAIR iteratively generates training data by identifying knowledge gaps, expanding evidence through API guidance, and refining distinctions via hard negative mining, thereby grounding retrieval in factual reality. Experiments show REPAIR surpasses 19 established baselines across nine materials science and biomedical benchmarks, underscoring the importance of factually augmenting data for long-tail deficits. AI

IMPACT Enhances the reliability of AI-driven scientific literature search and analysis tools.

RANK_REASON The cluster describes a new research paper detailing a novel framework for improving scientific information retrieval.

Read on Hugging Face Daily Papers →

• alphaXiv
• arXiv
• CatalyzeX Code Finder for Papers
• Connected Papers
• CORE Recommender
• DagsHub
• Gotit.pub
• Hugging Face
• Influence Flower
• Litmaps
• REPAIR
• ScienceCast
• scite Smart Citations

• paper
• infra

AI-generated summary · Google Gemini · from 2 sources. How we write summaries →