Reproduce “Finding the Right Balance: Relevance and Diversity in LLM Retrieval” 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
Guillaume Brouillette, Faustin Kagabo, Usef Faghihi, Nadia Ghazzali
The authors introduce a query-adaptive retrieval rule for questions needing distinct evidence from multiple passages. It estimates how many distinct embeddings the nearest-neighbor selections contain and diversifies when that estimate falls below the query’s evidence requirement. On a controlled HotpotQA test with heavy redundancy, tuned MMR gained 22.2 S-Recall@5 points over nearest-neighbor retrieval; this result applies to the injected condition, not universally.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.