BM25Okapi

BM25Okapi, often shortened to BM25, is a ranking function used by search engines to estimate the relevance of documents to a given search query. It is a probabilistic model that considers the term frequency within a document, the inverse document frequency of the terms across a corpus, and a document length normalization factor. The "Okapi" in its name refers to the Okapi information retrieval system developed at the City University London, where the BM25 family of functions was first introduced.

The core idea behind BM25 is that a document is more relevant if it contains the query terms frequently, if the query terms are rare in the overall collection of documents, and if the document is not excessively long compared to the average document length. It uses parameters, typically denoted as k1 and b, to control the impact of term frequency and document length, respectively. These parameters are often tuned for specific datasets or query types.

BM25 is a predecessor to more complex learning-to-rank models but remains a strong baseline and is widely used in practice due to its effectiveness and computational efficiency. Its probabilistic foundation provides a clear theoretical justification for its ranking behavior. The function calculates a score for each document in relation to a query, and documents are then presented to the user in descending order of these scores.