Research Article Recommender

Table of Contents

• Overview
• Installation
• Usage
• Contributing
• Acknowledgements
• License

Overview

This research-article recommender is a collaborative recommender that uses both a K-nearest neighbours algorithm and a matrix-factorization algorithm to recommend research articles based on an author's citation history.

Article data is obtained using the CORE API. Citation data is obtained using the OpenCitations API. An author-article matrix is generated using citation counts, i.e., the number of times an author has cited a particular article serves as an effective rating for that article. Citation count is perhaps the simplest way to turn citation history into a ratings system, but more complex relationships are certainly worth exploring. In particular, in this work, I included a citation decay factor in the rating system, essentially giving a citation a half-life of 25 years. This is intended to level the playing field between new and old articles.

Using Surprise, author-article citation ratings are fitted to both a KNNWithMeans K-nearest neighbours model and an SVG matrix-factorization model. For a particular author, the top-ranked articles from each model that the author has never cited are combined to yield a recommended collection of research articles.

The application is deployed as a (local) web app, shown in the following screenshots.

Installation

The backend of this app was written in Python and is served with Flask.

Start by cloning the repo:

$ git clone git@github.com:harnettd/research-article-recommender.git
$ cd research-article-recommender

To setup the needed Python environment, both requirements.txt and environment.yml are provided, the former for use with pip and the latter for use with conda. If using pip, setup the environment with

$ python -m venv venv
$ source venv/bin/activate
$ pip install -r requirements.txt

If using conda, setup the environment with

$ conda env create -f environment.yml
$ conda activate capstone

Next, set the article search parameters by editing src/search.py:

# Search the CORE API for these terms (there's an implied OR) 
# in the title, abstract, and full text of the article (if available).
SEARCH_TERMS = ['diquark', 'tetraquark', 'pentaquark']

# Search for articles published on or after this year.
YEAR = 2000

# The maximum number of articles to be found. Note that 10,000 is
# the upper limit here without a special ID needed for larger
# queries.
LIMIT = 1000

You will need API tokens for the CORE API and the OpenCitations API. Save the keys to the environment variables APIKEY_CORE and APIKEY_OPENCITATIONS respectively. Then, run the setup scipt:

$ source setup.sh

Usage

Change to the src/ directory and start the server:

$ cd src
$ flask run

Then, open index.html in a browser window. Enter the name of an author in the text input and click the SUBMIT button. To see a list of the authors that are contained in the data set, point the browser to http://127.0.0.1:5000/authors. To see a list of the articles that are in the dataset (as document identifier objecs or DOIs), point the browser to http://127.0.0.1:5000/dois.

Contributing

If you're interested in contributing to this project, please fork and clone the repo, write some code, and submit a pull request.

Acknowledgements

I'd like to thank the mentors and staff at Lighthouse Labs for their help and guidance throughout my data science bootcamp experience.

Licence

This project is licensed under the MIT License. See LICENSE for details.