Project configuration

Annif projects are used to set up backends and configure them with a specific vocabulary and parameters. Projects are defined in a file commonly called projects.cfg. By default, Annif looks for this file in the current directory where it is executed, but you can specify another path using the ANNIF_PROJECTS environment variable.

Here is an example project configuration. The configuration file format follows the INI style and each project is represented as a section, with the identifier of the project as the section name.

[tfidf-en]
name=TF-IDF English
language=en
backends=tfidf
analyzer=snowball(english)
limit=100
vocab=yso-en

A project has the following attributes:

Setting	Description
identifier	An identifier consisting of alphanumeric characters and basic punctuation.
name	A human-friendly name.
language	IETF BCP 47 language code.
backends	The backend (algorithm) that the project uses. See below for details.
analyzer	The analyzer used to pre-process and tokenize text. See below for details.
limit	The maximum number of results (subjects/concepts) to return
vocab	An identifier for the vocabulary used by this project

Some backends also require additional parameters (tfidf doesn't).

Backends

In most cases the backends attribute will contain just a single backend, for example tfidf or fasttext. However, it is possible to define a comma-separated list of backends whose results will be combined, like this:

tfidf,fasttext

Furthermore, you may give weights to each backend by suffixing the backend id with a colon and a weight value:

tfidf:1,fasttext:2

This way the results will be combined using a weighted average. In the above case the results from the fasttext backend will be weighted so that they are twice as important as the ones from tfidf. When not specified, the default weights for a backend is 1.

The ensemble backend provides a more general way of combining results from multiple backends.

Analyzers

Analyzers are used to pre-process, tokenize and normalize text. At the moment, Annif supports just two analyzers: simple and snowball.

The simple analyzer only splits text into words and turns them all into lowercase.

The snowball analyzer additionally performs stemming. It takes a language name as parameter, e.g. snowball(english) or snowball(finnish). You can use any language supported by the NLTK Snowball stemmer; see the NLTK stemmer documentation for details on supported languages.

Vocabularies

Most backends require a subject vocabulary. A vocabulary itself doesn't need any configuration, but the project must define an identifier for the vocabulary it uses. All projects with the same vocabulary identifier will share the same vocabulary, so the vocabulary only needs to be loaded (using the loadvoc command) for one of them.

Note that in Annif, vocabularies, like projects, are monolingual. If you load a multilingual SKOS vocabulary for a project, only the labels in the language defined for the project will be loaded.