Adeft

Adeft (Acromine based Disambiguation of Entities From Text context) is a utility for building models to disambiguate acronyms and other abbreviations of biological terms in the scientific literature. It makes use of an implementation of the Acromine algorithm developed by the NaCTeM at the University of Manchester to identify possible longform expansions for shortforms in a text corpus. It allows users to build disambiguation models to disambiguate shortforms based on their text context. A growing number of pretrained disambiguation models are publicly available to download through adeft.

Installation

Adeft works with Python versions 3.5 and above. It is available on PyPi and can be installed with the command

$ pip install adeft

Adeft's pretrained machine learning models can then be downloaded with the command

$ python -m adeft.download

Using Adeft

A dictionary of available models can be imported with from adeft import available_models

The dictionary maps shortforms to model names. It's possible for multiple equivalent shortforms to map to the same model.

Here's an example of running a disambiguator for ER on a list of texts

from adeft.disambiguate import load_disambiguator

er_dd = load_disambiguator('ER')

    ...

er_dd.disambiguate(texts)

Users may also build and train their own disambiguators. See the documention for more info.

Documentation

Documentation is available at https://adeft.readthedocs.io

Jupyter notebooks illustrating Adeft workflows are available under notebooks:

• Introduction
• Model building

Testing

Adeft uses nosetests for unit testing, and is integrated with the Travis continuous integration environment. To run tests locally, make sure to install the test-specific requirements listed in setup.py as

pip install adeft[test]

and download all pre-trained models as shown above. Then run nosetests in the top-level adeft folder.

Funding

Development of this software was supported by the Defense Advanced Research Projects Agency under award W911NF018-1-0124 and the National Cancer Institute under award U54-CA225088.