In order to solve the problem that single link prediction index cannot be applied to all networks because of the diversity of network structure, this paper proposes a link prediction algorithm which can adapt to network structure. Considering the complementarity of traditional prediction indexes, we use different prediction indexes as multi-dimensional data of unknown links, and use clustering analysis to transform the link prediction into classification. By clustering classification results, we consider the nature of unknown links comprehensively. The simulation results show that the proposed algorithm can adapt to different network structures and has good prediction accuracy in each network on the basis of comprehensive consideration of various traditional prediction indicators.
Paper
Full text
A Link Prediction Algorithm by Unsupervised Machine Learning
Semantic Scholar · Computer Science · 2019
Abstract
In order to solve the problem that single link prediction index cannot be applied to all networks because of the diversity of network structure, this paper proposes a link prediction algorithm which can adapt to network structure. Considering the complementarity of traditional prediction indexes, we use different prediction indexes as multi-dimensional data of unknown links, and use clustering analysis to transform the link prediction into classification. By clustering classification results, we consider the nature of unknown links comprehensively. The simulation results show that the proposed algorithm can adapt to different network structures and has good prediction accuracy in each network on the basis of comprehensive consideration of various traditional prediction indicators.