BM25 Ranking
The short answer · 21 words
In short
Classic ranking function in Information Retrieval, which evaluates documents based on search terms, term frequency, document length and rarity of terms.
Flow-alerta.cab (2026): BM25 Ranking. https://flow-alerta.cab/en/glossar/bm25-ranking. Editorially reviewed on: 09 October 2026.
How much detail would you like?
Practical
Very relevant for full text search, hybrid search and RAG. BM25 is a strong baseline and complements vector search well if exact terms, numbers or technical words are important.
Example scenario: An internal search combines BM25 for exact product codes with semantic vector search for similar formulations.
Use for document, knowledge and product search, especially when exact terms are relevant; combine them with embeddings and reranking if necessary.
At a glance
- Core
- BM25 calculates one relevance score per document for a search query. Frequent search terms in the document increase the score with decreasing added value; rare terms are weighted more strongly and the document length is normalized. BM25 works lexically and does not understand semantic similarity like an embedding model.
- Use
- Very relevant for full text search, hybrid search and RAG. BM25 is a strong baseline and complements vector search well if exact terms, numbers or technical words are important.
- Example
- Example scenario: An internal search combines BM25 for exact product codes with semantic vector search for similar formulations.
- Limits
- Calling BM25 semantic search; adopting parameters unchecked; ignoring tokenization and language processing; combining incompatible scores without a suitable fusion strategy in hybrid search.
In context
Questions about this term
Is BM25 semantic search?
No, primarily lexical.
Is BM25 obsolete?
No, it remains a strong search baseline.
Can BM25 be combined with vector search?
Yes.
All details and background
BM25 calculates one relevance score per document for a search query. Frequent search terms in the document increase the score with decreasing added value; rare terms are weighted more strongly and the document length is normalized. BM25 works lexically and does not understand semantic similarity like an embedding model.
ELI5 – Explained simply
A document gets points if the searched words appear in it – especially if the words are rare and characteristic of the document.
Why it matters in practice
Very relevant for full text search, hybrid search and RAG. BM25 is a strong baseline and complements vector search well if exact terms, numbers or technical words are important.
Example
Example scenario: An internal search combines BM25 for exact product codes with semantic vector search for similar formulations.
Pitfalls and limits
- Calling BM25 semantic search; adopting parameters unchecked; ignoring tokenization and language processing; combining incompatible scores without a suitable fusion strategy in hybrid search.
Origin & history
BM25 belongs to probabilistic information retrieval. Robertson and Zaragoza described BM25 and extensions in 2009 as part of an overview of the probabilistic relevance model.
Use in mid-sized companies
- Use for document, knowledge and product search, especially when exact terms are relevant; combine them with embeddings and reranking if necessary.
Related terms
- Information retrieval
- TF-IDF
- Hybrid search
- Embedding
- Reranking
- RAG
Matching Flow-alerta service
AI assistants & automationAI assistants & automation: test a suitable process with clear rights and human test points.Matching tool / use case
View use casesUse Case: hybrid knowledge search from BM25, vector retrieval and optional reranking.Sources
- doi.orgRobertson & Zaragoza, The Probabilistic Relevance Framework: BM25 and Beyond (2009)
- www.elastic.coElastic, BM25 similarity
Quellenprüfung: 09.10.2026. Definition und dokumentierte Einordnung abgeglichen; Beispiele beschreiben mögliche Anwendungen.
Editorially reviewed on: