Network Representation Learning

This project reproduces and analyzes five popular Unsupervised Network Representation Learning (UNRL) methods for graphs, focusing on node classification and link prediction tasks.

About

We implemented DeepWalk, Node2Vec, NetMF, LINE-1, and GraphSAGE to benchmark their performance on ten real-world network datasets. Our main focus was to reproduce results from Khosla et al. (2019) and to analyze outcomes when running these methods under resource constraints and on additional datasets.

Experimental Setup

• Evaluated models on social, citation, and collaboration networks (including BlogCatalog, Flickr, YouTube, Twitter, Cora, PubMed, DBLP-Ci, and Facebook Page-Page).
• Downstream tasks included node classification (using logistic regression and F1 scores) and link prediction (using ROC-AUC).
• Key implementation modifications included tuning hyperparameters for limited compute resources and extending experiments to the Facebook Page-Page dataset.
• All code was written from scratch using NetworkX, Gensim, scikit-learn, TensorFlow, and StellarGraph.

Resources

• Full project report: Adv_ML_Project.pdf

References

• Khosla, M., Setty, V., & Anand, A. (2019). "A comparative study for unsupervised network representation learning." IEEE Trans. on Knowledge and Data Engineering.
• Perozzi, B., Al-Rfou, R., & Skiena, S. (2014). "Deepwalk: Online learning of social representations." KDD.
• Řehůřek, R., & Sojka, P. (2010). "Software Framework for Topic Modelling with Large Corpora." LREC.
• Grover, A., & Leskovec, J. (2016). "node2vec: Scalable feature learning for networks." KDD.
• Qiu, J., et al. (2018). "Network embedding as matrix factorization." WSDM.
• Tang, J., et al. (2015). "LINE." WWW.
• Hamilton, W. L., Ying, R., & Leskovec, J. (2017). "Inductive representation learning on large graphs." NeurIPS.
• Rozemberczki, B., Allen, C., & Sarkar, R. (2019). "Multi-scale Attributed Node Embedding." arXiv:1909.13021.
• Hagberg, A., Swart, P., & S Chult, D. (2008). "Exploring network structure, dynamics, and function using NetworkX." Los Alamos National Lab.
• Data61 CSIRO. StellarGraph Library. https://github.com/stellargraph/stellargraph

Acknowledgments

• Developed as part of Advanced Machine Learning at KTH Royal Institute of Technology.
• Contributors: Adhithyan Kalaivanan, Aishwarya Ganesan, Daniel Richards, Vishal Nedungadi
• Forked from original repo: https://github.com/dannyrichy/graph-ml-project.git