jammywolf / machine-learning-with-graphs

1. Introduction

This repo summarizes papers I've read for machine learning on graphs. I'm also writing tutorials on zhihu.com and they're in Chinese.

2. Requirements

I use basic packages from Anaconda3 with Python 3.8.5. To make my life easier, I also use the following packages to implement models. Please see requirements.txt for the full list.

torch==1.7.0
torch_geometric==1.6.3
ogb==1.2.3
scikit-multilearn==0.2.0

3. Papers

The following are papers that I'll cover in this repo.

3.1 Early Research

3.1.1 Factorization-Based Models

• Distributed large-scale natural graph factorization. Amr Ahmed, Nino Shervashidze, Shravan Narayanamurthy, Vanja Josifovski, and Alexander J Smola. WWW 2013.

• Grarep: Learning graph representations with global structural information. Shaosheng Cao, Wei Lu, and Qiongkai Xu. CIKM 2015.

• Asymmetric transitivity preserving graph embedding. Mingdong Ou, Peng Cui, Jian Pei, Ziwei Zhang, and Wenwu Zhu. KDD 2016.

3.1.2 Random Walk-Based Models

• Deepwalk: Online learning of social representations. Bryan Perozzi, Rami Al-Rfou, and Steven Skiena. KDD 2014.

• node2vec: Scalable feature learning for networks. Aditya Grover and Jure Leskovec. KDD 2014.

• struc2vec: Learning node representations from structural identity. Leonardo FR Ribeiro, Pedro HP Saverese, and Daniel R Figueiredo. KDD 2017.

3.1.3 GCN-based Models

3.2 Scalability and Expressivity

3.2.1 Node Sampling

3.2.2 Subgraph Sampling

3.2.3 Regularization

3.2.4 Architecture

3.3 Incorporating Edge and Label Information

3.3.1 Incorporating Edge Information

3.3.2 Incorporating Label Information

3.4 Training Strategy

3.5 Generalization to Heterogeneous Graphs

3.5.1 Random Walk-Based Models

3.5.2 GCN-Based Models

3.5.3 Application

3.6 Interpretability and Theory Guidance

3.6.1 Expressive Power of GCNs

3.6.2 When Will GCNs Fail

3.6.3 How to Design Better GCNs