Stacking models for nearly optimal link prediction in complex networks
Abstract
Networks are a powerful tool for modeling complex biological and social systems. However, most networks are incomplete, and missing connections can negatively affect scientific analyses. Today, many algorithms can predict missing connections, but it is unknown how accuracy varies across algorithms and networks and whether link predictability varies across scientific domains. Analyzing 203 link prediction algorithms applied to 550 diverse real-world networks, we show that no predictor is best or worst overall. We then combine these many predictors into a single state-of-the-art algorithm that achieves nearly optimal performance on both synthetic networks with known optimality and real-world networks. Not all networks are equally predictable, however, and we find that social networks are easiest, while biological and technological networks are hardest.
Document Details
- Document Type
- Pub Defense Publication
- Publication Date
- Sep 04, 2020
- Source ID
- 10.1073/pnas.1914950117
Entities
People
- Aaron Clauset
- Amir Ghasemian
- Aram Galstyan
- Edoardo Airoldi
- Homa Hosseinmardi
Organizations
- Army Research Office
- Harvard University
- National Science Foundation
- Santa Fe Institute
- Temple University
- University of Colorado
- University of Southern California