Concept Representation Learning
We consider a graph of papers and concepts. Paper nodes have a title and are connected when they have at least one common author. Concept nodes are featureless. The task is to learn a representation for the featureless concept nodes.
make to setup a virtual environment, download sample data, and run the experiments.
If there are any issues, or the methods should be applied to other data, follow the detailed steps below.
Set up a virtual environment (i.e.
virtualenv -p /usr/bin/python3 venv && source venv/bin/activate) and then run:
pip install -r requirements.txt
In case, the required packages are not backwards compatible, run
pip install -r requirements-stable.txt instead.
A graph directory
graph_dir with three csv files:
- paper.csv with columns: paper_id, year, title
- annotation.csv with columns: paper_id, subject
- authorship.csv with columns: paper_id, author
python3 train.py gcn_cv_sc graph_dir -o model_dir
For more information on hyperparameters, consult
python3 main.py -h.
Then you can evaluate the resulting embedding
model_dir/embedding.csv with the scripts
make experiment to reproduce the experiments from the paper. This will download a sample of 100k research items to construct the graph.
If you are interested in contributing, please notify us.