/BMASS

Code for Newman-Griffis et al. (2017) "Insights into Analogy Completion from the Biomedical Domain"

Primary LanguagePythonMIT LicenseMIT

BMASS

Source code used for ACL BioNLP 2017 workshop paper:

Denis Newman-Griffis, Albert M Lai, Eric Fosler-Lussier. Insights into Analogy Completion from the Biomedical Domain

Dataset

Download the dataset (requires a valid UMLS Terminology Services login).

Code

  • analogy_task: implementation of the analogy task (using TensorFlow v0.7)
  • BMASS: parser for BMASS data files
  • lib: various dependencies

A demo virtual machine setup is also included in the demo directory, using Vagrant. This will run the analogy experiment for the full BMASS dataset on CBOW and skip-gram embeddings pre-trained on the 2016 PubMed baseline.

To run the demo:

cd demo
vagrant up
vagrant ssh
$ cd /vagrant/src
$ make demo

PubMed embeddings

The embeddings trained for this paper can be downloaded from here.