medium10 pts · first answer countsembeddingsvector-search

Dot product on unit vectors

All embeddings in your index are normalised to unit length. You switch the search metric from cosine similarity to dot product. What happens to the result ranking?

A.It reverses, because dot product prefers distant vectorsB.It stays the same, because dot product equals cosine for unit vectorsC.It becomes random, because dot product ignores directionD.It favours longer documents

Discussion 0

Join free to post in the discussion.

Dot product on unit vectors

Discussion 0

Related questions