Link Prediction
Definition
Link prediction estimates which edges are missing in a graph or may form in the future.
Methods use structure, node features, or graph embeddings. A high model probability does not mean the relationship actually exists.
Simply explained
How to predict in a network which two nodes are likely connected.
Why it matters in practice
Relevant for recommender systems, knowledge graphs, fraud networks, and relationship analysis.
Example from business
A knowledge graph suggests possible component-failure relationships for expert review.
Limits and pitfalls
- Treat unknown edges as negative; leakage via temporally later graph data; suggestions adopted unverified as facts.
Origin & history
Liben-Nowell and Kleinberg studied link prediction in social networks; a shortened version appeared in 2003, an author version is dated January 8, 2004.
Use in mid-sized companies
- Possible application in SMEs: generate candidate relationships and then validate them professionally.
Frequently asked questions
Does link prediction generate new facts?
No, only predictions. Which data is important? Graph structure and optionally node/edge features.
Sources
- The Link Prediction Problem · Liben-Nowell & Kleinberg (2007), The Link Prediction Problem
- Ergänzender Primärbeleg · Ergänzender Primärbeleg
Source checked on 10./12.10.2026.
Editorially reviewed on: