Uncovering the Structural Fairness in Graph Contrastive Learning.pptx
1. Van Thuy Hoang
Network Science Lab
Dept. of Artificial Intelligence
The Catholic University of Korea
E-mail: hoangvanthuy90@gmail.com
2023-11-26
Ruijia Wang; NeurIPS 23
2. 2
Graph Convolutional Networks (GCNs)
Generate node embeddings based on local network neighborhoods
Nodes have embeddings at each layer, repeating combine messages
from their neighbor using neural networks
4. 4
phenomenon
Graph Contrastive Learning (GCL)
GCL integrates the power of GCN and contrastive learning.
GCL relieves GCN from annotations, and displays SOTA performance in
many tasks
5. 5
Contributions
The first to discover that GCL methods exhibit more structural fairness than
GCN
Theoretically validate the reason for structural fairness in GCL is that GCL
stimulates intracommunity concentration and inter-community scatter
Propose a method GRADE to further improve structural fairness by enriching
the neighborhood of tail nodes while purifying neighbors of head nodes.
Comprehensive experiments demonstrate that our GRADE outperforms
baselines on multiple benchmark datasets and enhances the fairness to degree
bias.
6. 6
The Proposed Model
Aim to increase intra-community edges while decreasing inter-community
edges
Tail nodes:
Interpolate the ego network of the
anchor tail node with that of a similar
node
Head nodes:
Purify the neighborhood by similarity-
based sampling.
7. 7
Graph Augmentation
Topology Augmentation
The similarity matrix based on cosine similarity of representations
1) For any tail node v_tail, we sample a node v_sample
2) The similarity
Feature Augmentation
randomly sample a mask
8. 8
Optimization Objective
Node representations h_i and o_i from different graph augmentations form the
positive pair.
Node representations of other nodes in graph augmentations are regarded as
negative pairs.
The overall objective to be maximized is the average of all positive pairs
9. 9
Experiments
Datasets
1) citation networks including Cora and Citeseer
2) social networks Photo and Computer [23] from Amazon
Evaluation Protocol :
state-of-the-art GCL models
12. 12
Conclusion
Structural unfairness for node representation learning
Node representations obtained by GCL methods are fairer to degree bias than
those learned by GCN, and explore the underlying cause of this phenomenon.
Based on our theoretical analysis, we further propose a novel GCL model
targeting degree bias.
A limitation of GRADE is its heuristic design