Reproduce “Nothing Left Behind: Embedding Retrieval and Language Model Screening for High Recall Evidence Search” 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
Giuliano Martinelli, Abelardo Carlos Martinez Lorenzo, Satvik Garg, Linxi Wang, Riccardo Orlando, Piriyakorn Piriyatamwong, Samuel Paul Fraiberger
The authors separate evidence retrieval from organization by matching review questions to intervention–outcome study pairs, retrieving the K candidates, and screening them with a language model. Taxonomy labels can then group results. In ImpactAI, the method found 85.0% of relevant studies at K = 2,500, versus 45.1% for taxonomy retrieval, across 348 questions; recall covers only review studies present in ImpactAI.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.