Zum Hauptinhalt springen

Work through cosine similarity

Interpret cosine similarity as normalized dot product and use it to reason about ranking.

embeddings-explained5 min read

A query vector for "refund after renewal" is compared with two candidate chunks. Candidate A is a short refund FAQ. Candidate B is a long subscription policy page with refunds, billing, upgrades, and account ownership. Cosine similarity = normalized dot product: compare direction after accounting for vector length. The common trap is assuming the longest or most detailed document should rank highest. For semantic retrieval, the closest directional match may be a shorter chunk. Represent Turn the query and each candidate chunk into vectors in the same embedding space. A vector is a comparable numerical representation. You cannot compare text…

Read the full lesson

Sign up free: one personalized lesson every day, matched to your role and goals.

← Back to library

Kontakt aufnehmen