EasyDeepLearn

How does Reciprocal Rank Fusion (RRF) work?

medium

Answer

  • For each doc d, compute RRF score = sum over retrievers r of 1/(k  +  rankr(d))1 / (k\; + \;\operatorname{rank}_{r}(d)), with k=60 typically.
  • Rank-based (doesn't need score calibration between retrievers of different scales — critical since BM25 scores and cosine similarities are on different scales).
  • Robust, simple, no learned parameters.
  • Standard fusion for hybrid dense + BM25 search.
  • Alternative: learned fusion via a small cross-encoder on top-N candidates.
Check yourself — multiple choice
  • Sum raw scores
  • Rank-based fusion: score = Σ 1/(k  +  rankr)1 / (k\; + \;\operatorname{rank}_{r}) — no score calibration needed, simple, robust standard for hybrid
  • Only max score
  • Random pick

RRF: rank-based fusion (Σ 1/(k+rank)); avoids score-scale calibration issues.

#retrieval#hybrid-search

Practise LLMs & GenAI

214 interview questions in this topic.

Related questions