The New Link Prediction Methods Based on Spectral Analysis

摘要

The research of link prediction has been widely concerned. Most of the algorithms analyze the topology information of networks to judge whether there is any connection between nodes. Recently, the new link prediction indices about social networks need to know the additional attribute information of nodes, but it is very difficult and the practicality is not very high. In this paper, we propose two new indices SALP and AD-SALP that consider the properties of the links. The spectral analysis is introduced into the methods and the eigenvectors of Laplacian matrix are used to map the nodes to the two-dimensional space that it is easy to calculate the similarity distance between a pair of nodes. Then the similarity calculation of the nodes is directly transformed into the supervised binary classification prediction problem for the edges. The experimental results on six real-world networks prove the feasibility of the SALP index and show that the AD-SALP index outperforms other prediction baselines.