Learning Representations via Spectral-Biased Random Walks on Graphs

About

This is a source code for our IJCNN 2020 paper: Learning Representations using Spectral-Biased Random Walks on Graphs. We propose a novel mechanism which uses random walks biased by Spectral similarity of nodes. The proposed method enhance the SOTA results on many datasets of varying attributes.

Requirements

Please download miniconda and create an environment using the following command:

conda create -n pytorch35

Activate the environment before executing the program as follows:

source activate pytorch35

Installation

Run the following command to install the dependencies required to run the code.

python -r requirements.txt

If your default python version is 2.X, we request you to switch to 3.X.

Dataset

We used ten different datasets for Link Prediction Task and three datasets for Node Classification. The dataset names are:

• Power
• C.elegans
• USAir
• Road-Euro
• Road-Minnesota
• Bio-SC-GT
• Infect-hyper
• PPI
• Facebook
• HepTh

---

• Cora
• Citeseer
• PubMed

Usage

The directory End2End contains the end to end pipeline to run the code. Run the following command to reproduce the results on Celegans dataset.

bash end2end.sh

The bash file currently has code for Celegans dataset, provided as an example. You can change the dataset name and corresponding hyperparameters from the table given in the paper to reproduce the results. This directory has the complete pipeline, from generating the spectral walks, running the par2vec model to learn primary node embeddings and finally running the link prediction model.

The src directory contains the code just for link prediction and node classification with precomputed spectral-biased random walk and paragraph vector model output. Run the following command to run just the link prediction part of the entire pipleline

python LP.py

The LP.py is the python script which will run the link prediction model on all datasets and store the results in a text file inside the directory result_logs/LP.

The NC.py is the python script which will run the node classification model on the datasets- cora and citeseer and store the results in a text file inside the directory result_logs/NC.

Citation

Please cite the paper if you use this code.

@article{sharma2020learning,
  title={Learning Representations using Spectral-Biased Random Walks on Graphs},
  author={Sharma, Charu and Chauhan, Jatin and Kaul, Manohar},
  journal={arXiv preprint arXiv:2005.09752},
  year={2020}
}