dgkngk / ptcdb

Patents citation dataset that we generated for usage in gcn applications for our bachelor's finishing project.

ptcdb

This dataset is created as a citational graph dataset, to be used with a graph citational network implementation such as this. See Kipf & Welling's paper for more information about graph convolutional networks.

Usage

Under data batches file, there are 4 batches that we generated from USPTO patent database files. Under the example implementation file, there is an implementation of SVM and GCN(with PyTorch, from pygcn). This particular implementation used the Cora dataset, we modified it to use the patent citation dataset with the same format as the Cora's. We also added an SVM test to compare the two implementations with.

Setup

$ cd example_implementation
$ python setup.py install
$ cd pygcn
$ python train.py

You can also change the dataset back to Cora, under utils.py to record the difference in two datasets.