A bi-scale method of link prediction

Link prediction in networks is of both theoretical interest and practical significance in many branches of science, and a great number of algorithms are based on microscale (common neighbours) or mesoscale (communities) information of observed networks. Either microscale or mesoscale methods are limited in the understanding of the topological properties at the corresponding scales. This article proposes a bi-scale model for predicting missing links, which fuses the microscale (RA Index) and mesoscale (NC Index) methods. Empirical experiments show this bi-scale index enhances the advantage and avoids the disadvantage of single scale approaches at certain degree. As the experiments show, the proposed method on 7 out of 10 disparate real networks outperforms some mainstream link prediction baselines with 20% missing links, such as NC Index, CN Index, AA Index, RA Index, and so on.