Graph InfoClust: Leveraging cluster-level node information for unsupervised graph representation learning

15 Sep 2020 Costas Mavromatis George Karypis

Unsupervised (or self-supervised) graph representation learning is essential to facilitate various graph data mining tasks when external supervision is unavailable. The challenge is to encode the information about the graph structure and the attributes associated with the nodes and edges into a low dimensional space... (read more)

PDF Abstract

Results from the Paper


TASK DATASET MODEL METRIC NAME METRIC VALUE GLOBAL RANK RESULT BENCHMARK
Node Classification AMZ Comp Graph InfoClust (GIC) Accuracy 81.5% # 3
Node Classification AMZ Photo Graph InfoClust (GIC) Accuracy 90.4% # 4
Link Prediction Citeseer Graph InfoClust (GIC) AUC 97 # 1
AP 96.8 # 1
Node Classification Citeseer Graph InfoClust (GIC) Accuracy 71.9% # 28
Node Clustering Citeseer Graph InfoClust (GIC) Accuracy 69.6 # 1
NMI 45.3 # 1
ARI 46.5 # 1
Node Classification Coauthor CS Graph InfoClust (GIC) Accuracy 89.4% # 6
Node Classification Coauthor Phy Graph InfoClust (GIC) Accuracy 93.1 # 2
Node Clustering Cora Graph InfoClust (GIC) Accuracy 72.5 # 2
NMI 53.7 # 5
ARI 50.8 # 1
Link Prediction Cora Graph InfoClust (GIC) AUC 93.5% # 3
AP 93.3% # 3
Node Classification Cora: fixed 20 node per class Graph InfoClust (GIC) Accuracy 81.7 # 5
Node Clustering Pubmed Graph InfoClust (GIC) Accuracy 67.3 # 3
NMI 31.9 # 3
ARI 29.1 # 1
Node Classification Pubmed Graph InfoClust (GIC) Accuracy 77.4% # 39
Link Prediction Pubmed Graph InfoClust (GIC) AUC 93.7% # 5
AP 93.5% # 5

Methods used in the Paper


METHOD TYPE
GIC
Graph Embeddings