Graph Representation Learning:
Algorithms and Applications
Jian Tang
HEC Montreal
CIFAR AI Chair, Mila
Email: jian.tang@hec.ca
Social Networks
Facebook
Twitter
Graph from Albert-László Barabási’ s SIGIR09 keynote
3
Protein-Protein Interaction Graph
https://www.ebi.ac.uk/training/online/course/network-analysis-protein-interaction-data-introduction/protein-protein-interaction-networks
Drug-Protein Interaction Graph
https://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1002503
Knowledge Graphs
• Multiple types of edges
• Each corresponds to a relation type
• A set of facts, each of which is represented as a triplet
• (Bill_Gates, CoFounder, Microsoft)
friend?
Under review as a conference paper at ICLR 2020
Molecules
O
F
NH
S
N
Br
N
NH
I
N
NH
Cl
O
N
HO
N
N
Cl
HO
NH
Cl
Cl
O
F
S
O
O
NH2
N
O
HO
O
N
N
H2N
Cl
SH
O
NH
OH
NH
Br
N
S
Cl
O
Br
Cl
NH
O
H2N
NH
N
N
N
N
N
O
NH
F
N
O
O
O
O
I
O
N
NH
O
H2N
N
O
Br
O
I
N
S
OH
O
N
O
N
NH
NH
N
N
O
NH2
O
N
NH
O
NH
Various Applications on Graphs
• Predicting whether a user is a democratic or republican in Facebook?
• Recommending friends in social networks
• Predicting missing facts on knowledge graphs
• Predicting the effective drugs for a target disease in a biomedical
knowledge graph, a.k.a. drug repurposing
• Predicting the chemical properties of molecules
• …
• Most of these applications require good feature representation of
graphs!!