Mining relationships between performance of link prediction algorithms and network structure

The numerous link prediction algorithms proposed by the network science researchers demonstrate their creativity in this hot topic. However, various algorithms together with the miscellaneous real-world networks put much difficulty on the choice of algorithm when coping with a new network. In this paper, we try to provide some elementary rules through mining the relationships between network structure features and the algorithm mechanisms. We discovered some principles indicating clustering coefficients influences on the prediction accuracy of structure-based algorithms. Besides, our experiment results present some interesting phenomenon neglected previously. The results and discussions may help us understand the link prediction problem better and further. (c) 2021 Elsevier Ltd. All rights reserved.

Automated summary

This paper provides some elementary rules through mining the relationships between network structure features and the algorithm mechanisms, and discovers the impact of clustering coefficients on the prediction accuracy. Experimental results present some interesting phenomena which offer new insights for solving the link prediction problem.