Build Status Coverage Status

AnyStyle is a very fast and smart parser for academic references. It was originally inspired by ParsCit and FreeCite; AnyStyle uses machine learning algorithms and aims to make it easy to train the model with data that is relevant to your parsing needs.

Using AnyStyle CLI

$ [sudo] gem install anystyle-cli
$ anystyle --help
$ anystyle help find
$ anystyle help parse

See anystyle-cli for more details.

Using AnyStyle in Ruby

Install the anystyle gem.

$ [sudo] gem install anystyle

Once installed, you can use the static Parser and Finder instances by calling the AnyStyle.parse or AnyStyle.find methods. For example:

require 'anystyle'

pp AnyStyle.parse 'Derrida, J. (1967). L’écriture et la différence (1 éd.). Paris: Éditions du Seuil.'
#-> [{
#  :author=>[{:family=>"Derrida", :given=>"J."}],
#  :date=>["1967"],
#  :title=>["L’écriture et la différence"],
#  :edition=>["1"],
#  :location=>["Paris"],
#  :publisher=>["Éditions du Seuil"],
#  :language=>"fr",
#  :scripts=>["Common", "Latin"],
#  :type=>"book"

Alternatively, you can create your own AnyStyle::Parser or AnyStyle::Finder with custom options.

Using the AnyStyle Web App

AnyStyle is available as web application at

The web application is open source and you can also host yourself!

Improving results for your data


You can train custom Finder and Parser models. To do this, you need to prepare your own data sets for training. You can create your own data from scratch or build on AnyStyle's default sets. The default parser model is based on the core data set; the default finder model source data is not publicly available in its entirety, but you can find a number of tagged documents here.

When you have compiled a data set for training, you will be ready to create your own model:

$ anystyle train training-data.xml custom.mod

This will save your new model as custom.mod. To use your model instead of AnyStyle's default, use the -P or --parser-model flag and, respectively, -F or --finder-model to use a custom Finder model. For instance, the command below would parse all references in bib.txt using the custom model we just trained and print the result to STDOUT using the JSON output format:

$ anystyle -P custom.mod -f json parse bib.txt -

When training your own models, it is good practice to check the quality using a second data set. For example, using AnyStyle's own gold data set (a large, manually curated data set) we could check our custom model like this:

$ anystyle -P x.mod check ./res/parser/gold.xml
Checking gold.xml.................   1 seq  0.06%   3 tok  0.01%  3s

This command will print the sequence and token error rates; in the case of AnyStyle a the number of sequence errors is the number of references which were tagged differently by the parser than they were in the input; the number of token errors is the total number of words across all the references which were tagged differently. In the example above, we got one reference wrong (out of 1700 at the time); but even this one reference was mostly tagged correctly, because only a total of 3 words were tagged differently.

When working with training data, it is a good idea to use the Wapiti::Dataset API in Ruby: it supports all the standard set operators and makes it very easy to combine or compare data sets.

Natural Languages used in AnyStyle

As mentioned above, the core dataset contains the manually marked-up references that are used as the basis for the default AnyStyle parsing model. If the references you are trying to parse include many non-English documents, the distribution of natural languages in this corpus is relevant (detected using cld).

Language n
Others 9
Not reliably determined 449
(but mainly English)

(These data are based on AnyStyle version 1.3.13)

There is a strong prevalence of English-language documents with the conventions used in English-language bibliographies, with some representation of other European languages. The languages used reflect those used in scientific publishing as well as the maintainers' competencies. If you are working with many documents in languages other than English, you might consider training the model with some examples in the relevant languages.

AnyStyle should work with references written in any Latin script (including most European languages, languages such as Indonesian and Malaysian, as well as romanised Arabic, Chinese and Japanese). It should also support languages written with non-Latin alphabets (such as Russian), although no examples of these appear in the default training sets. Languages written in syllabaries or complex symbols which do not use white space to separate tokens are not compatible with AnyStyle's approach: this includes Chinese, Japanese, Arabic as well as many Indian languages.

Dictionary Adapters

During the statistical analysis of reference strings, AnyStyle relies on a large feature dictionary; by default, AnyStyle creates a persistent Ruby Hash in the folder of the anystyle-data Gem. This uses up about 2MB of disk space and keeps the entire dictionary in memory. If you prefer a smaller memory footprint, you can alternatively use AnyStyle's GDBM dictionary. GDBM bindings are part of the Ruby standard library and are supported on all platforms, but you may have to install GDBM on your platform before installing Ruby.

If you do not want to use the the persistent Ruyb Hash nor the GBDM bindings, you can store your dictionary in memory (not recommended) or use a Redis. The best way to change the default dictionary adapter is by adjusting AnyStyle's default configuration (when using the default parser instances you must set the default before using the parser):

AnyStyle::Dictionary.defaults[:adapter] = :ruby
#-> Use a persistent Ruby hash;
#-> slower start-up than GDBM but no extra dependency

AnyStyle::Dictionary.defaults[:adapter] = :hash
#-> Use in-memory dictionary; slow start-up but uses no space on disk

require 'anystyle/dictionary/gdbm'
AnyStyle::Dictionary.defaults[:adapter] = :gdbm

To use Redis, install the redis and redis/namespace (optional) Gems and configure AnyStyle to use the Redis adapter:

AnyStyle::Dictionary.defaults[:adapter] = :redis

# Adjust the Redis-specifi configuration
require 'anystyle/dictionary/redis'
AnyStyle::Dictionary::Redis.defaults[:host] = 'localhost'
AnyStyle::Dictionary::Redis.defaults[:port] = 6379

About AnyStyle


The AnyStyle source code is hosted on GitHub. You can check out a copy of the latest code using Git:

$ git clone

If you've found a bug or have a question, please open an issue on the AnyStyle issue tracker. Or, for extra credit, clone the AnyStyle repository, write a failing example, fix the bug and submit a pull request.


AnyStyle is a volunteer effort and we encourage you to join us! Over the years our main contributors have been:


Copyright 2011-2023 Sylvester Keil. All rights reserved.

AnyStyle is distributed under a BSD-style license. See LICENSE for details.