11 Chapter

Graph & Network ML

Learn from data defined over nodes and edges.

Graph and network ML learns from data defined over nodes and edges — rankings, communities, and link structure. The methods below span classic centrality and community detection through to modern graph neural networks for representation learning.

  • Use PageRank for node importance and Louvain/Leiden for community detection.
  • Use Graph Neural Networks (GraphSAGE/GCN/GAT) when you must learn from graph structure.
#AlgorithmBest forCommon fields
1PageRank Node importance
  • Search
  • citation networks
  • fraud signals
2Community Detection: Louvain, Leiden Finding groups
  • Social networks
  • biology
  • fraud rings
3Node2Vec / DeepWalk Graph embeddings
  • Recommendations
  • link prediction
4Graph Neural Networks Learning from graph structure
  • Molecules
  • fraud
  • knowledge graphs
5GraphSAGE / GCN / GAT Scalable graph representation learning
  • Social networks
  • recommender systems
  • cybersecurity
6Link Prediction Models Predicting relationships
  • Knowledge graphs
  • social networks
  • drug discovery