Tool for automated subject indexing and classification
Tool for automated subject indexing and classification
Choose a controlled subject vocabulary and train Annif on already indexed documents – it can then suggest subjects for new documents!
Annif uses a combination of existing natural language processing and machine learning tools including TensorFlow, Omikuji, fastText and Gensim. It is multilingual and can support any subject vocabulary (in SKOS or a simple TSV format). It provides a command-line interface, a simple Web UI and a microservice-style REST API.
Code and documentation for Annif is available on GitHub (Apache 2.0 license). Annif can also be installed from PyPI and as a Docker image from Quay.io. Annif is mainly being developed at the National Library of Finland, but others are welcome to join in!
There is a collection of downloadable Annif models in the 🤗 Hugging Face Hub.
To get a hands-on experience of Annif, study the Annif tutorial materials, which include example data sets, exercises and short video presentations:
There is also extensive usage documentation in the wiki on GitHub.
The annif-users mailing list and web forum is available on Google Groups. The forum is meant for general discussion about Annif, asking for help, and announcements of new versions. All messages are public and anyone is welcome to join!
Please use the forum instead of sending personal e-mail to the Annif developers.
Finto AI - service for automated subject indexing.
Yle, the Finnish Broadcasting Company, uses Annif to assign tags to online news articles.
The German National Library uses Annif as the core of its automated subject indexing system Erschließungsmaschine (EMa).
The National Library of Poland uses Annif as a part of the DESKRYPTOR service for automated subject indexing.
AATA, a research database by the Getty Conservation Institute, uses Annif in their indexing process to suggest terms from Getty's Art and Architecture Thesaurus.
Storia Oy generates metadata about upcoming books with Annif.
In Jyväskylä University Digital Repository and in repositories of other institutes (Osuva, Trepo, Theseus, Taju, Lauda) Annif assists the subject indexing of theses and dissertations.
ZBW – The Leibniz Information Centre for Economics uses Annif as a part of their automated indexing service AutoSE (read more here).
More users of Annif and/or Finto AI
A pre-print is available on arXiv that describes the Annif system which was used in participating to the SemEval-2025 Task 5: LLMs4Subjects. Our system was ranked 🥇 1st in the category where the full vocabulary was used, 🥈 2nd in the smaller vocabulary category and 🏅 4th in the qualitative evaluations.
The Annif user survey report (2025) includes opinions about Annif as well as insights and challenges related to its use.
An article that investigates the usage of Annif for Dewey Decimal Classification was published in 2024 in the Journal of Documentation.
An article about Annif and Finto AI has been published in 2022 in the peer-reviewed Open Access journal JLIS.it.
Annif Analyzer Shootout paper was also published in 2022 in Code4Lib.
The first article about Annif was published in 2019 in LIBER Quarterly.
The software itself is also archived on Zenodo and has a citable DOI. See the README on the Annif GitHub project site for more details including BiBTeX snippets.
Above is the presentation of Annif and Finto AI at SWIB20.
See also the SWIB18 presentation, MDN Workshop 2020 presentation, and the video of the first prototype of Annif.
