Information retrieval

Similarity

How a ranker decides that two items are close: a lexical score such as BM25, or a vector score such as cosine. I study the choice, and how to tell that a new score did not quietly get worse.

Broader subject: Relevance