This repository hosts code for the papers:

• On the utilization of structural and textual information of a scientific knowledge graph to discover future research collaborations: a link prediction perspective (Discovery Science - DS 2020) - Download

Available in this link

Edit experiments.ipynb.

Prequisites:

• Windows 10 64-bit / Debian based Linux 64-bit.
• Python 3 (min. version 3.6), pip3 (& py launcher Windows-only).
• Working Neo4j Database (min. version 4.2.1).

Download the project from the green button above, unzip it,
and then open a cmd terminal to this folder and type pip3 install -r requirements.txt.
This command will install the neccessary Python libraries* to run the project.

We ran the following commands to update Python, git,
clone the project to a local folder and install the necessary Python libraries*.

sudo apt install python3.6
sudo apt install git-all
git clone https://github.com/NC0DER/CORD19_GraphOfDocs
cd CORD19_GraphOfDocs
pip3 install -r requirements.txt

* Optionally you could create a virtual environment first,
* to isolate the libraries from your python user install.
* However the setup script doesn't downgrade existing libraries,
* so there's zero risk in affecting your local user install.

Create a new database from the Neo4j desktop app using 4.2.1 as the min. version.
Update your memory settings to match the following values,
and install the Graph Data Science Library plugin as depicted in the image. Hint: if you use a dedicated server that only runs Neo4j, you could increase these values, accordingly as specified in the comments of these parameters.

Run the CORD19_GraphOfDocs.py script which will create thousands of nodes, and millions of relationships in the database.
Once it's done, the database is initialized and ready for use.

You could use the Neo4j Browser to run your queries, or for large queries you could use the custom visualization tool
visualize.html which is located in the CORD19_GraphOfDocs Subdirectory.

• Nikolaos Giarelis ([email protected])
• Nikos Kanakaris ([email protected])
• Nikos Karacapilidis ([email protected])