hard20 pts · first answer countshybrid-searchbm25

Adding BM25 and cosine scores

A teammate merges keyword and vector search with final = 0.5 × BM25 score + 0.5 × cosine score. Results look dominated by keyword matches. Why?

A.BM25 scores are unbounded and dwarf cosine values; normalise or fuse by rankB.Cosine scores are always negative for the most relevant documentsC.BM25 scores cannot be combined with any other kind of retrieverD.The weights must add to more than 1 for hybrid search to work

Discussion 0

Join free to post in the discussion.

Adding BM25 and cosine scores

Discussion 0

Related questions