ucla-bd2k / aztec_pub_classifier Goto Github PK

Find the patterns of sentences that tell us that the tool is available.

As a developer, I'd like to use a neural network to classify publications.

Comments to code
High level wiki on Github on: how to setup and run.

Old publications will have links that are broken

given a sentence, classify it as important or not important.
Important: says that the tool is accessible
Not important: everything else

Some tools will have multiple publications, possibly for different versions

Try adding new features to classifier, e.g.:

• presence of colon in title
• case changes in title
• MeSH terms

Using the Pubmed API, classify the publications based on the title and abstract of the publications.

Example usage of API:
Gets a list of ID's of the publications: (returns 1000, set the parameter accordingly)
http://eutils.ncbi.nlm.nih.gov/entrez/eutils/esearch.fcgi?db=pubmed&retmode=json&retmax=1000&sort=relevance&term=%22Bioinformatics%22&field=journal
Get's information about the particular publication (change the id parameter accordingly)
https://eutils.ncbi.nlm.nih.gov/entrez/eutils/efetch.fcgi?db=pubmed&id=27402908&rettype=xml

Tools that have Availability in the abstract are tools.

Include context as well

As a developer, I would like to understand how to use neural networks to classify publications based on the text.

Tutorial: http://jrmeyer.github.io/tutorial/2016/02/01/TensorFlow-Tutorial.html

• look at data to manually determine which words or phrases determine tool or not tool
• use list of programming languages to create feature that indicates presence of a programming language
• create classifier that only uses above features and presence of colon, capital letters in title, and not total word freqencies