Keywords in OpenAlex are short phrases that describe what a work is about. There are over 26,000 keywords in the system, and each work can be tagged with up to 5 of them. Keywords are derived from topics — each of the ~4,500 topics has 10 associated keywords.
How keywords are assigned
The process for assigning keywords to a work has four steps:
- Gather candidate keywords — Take the topics assigned to the work (up to 3) and pull the keywords associated with those topics. This gives up to 30 candidate keywords.
- Score similarity — Score each candidate keyword's similarity to the work's title and abstract using embeddings from the BGE M3-Embedding model. This is a multilingual embedding model that captures semantic meaning, so it can match keywords even when the exact phrase doesn't appear in the text.
- Apply a threshold — Only keywords that score above a similarity threshold are kept. This filters out keywords that are related to the work's topic area but aren't really relevant to the specific work.
- Keep the top 5 — From the keywords that pass the threshold, the top 5 by similarity score are assigned to the work.
Source code
The keyword extraction pipeline is open source: openalex-keywords (v2) on GitHub.
API access
You can browse and filter keywords through the API:
https://api.openalex.org/keywords
https://api.openalex.org/keywords/type-1-diabetesTo find works with a specific keyword:
https://api.openalex.org/works?filter=keywords.id:machine-learningFor full details, see the Keywords API documentation.
If you see any issues with keywords, feel free to submit a support request.