Predicting Missing Links of Complex Network via Effective Common Neighbors

Abstract

Link prediction can predict the missing links of complex networks, which promotes a better understanding of evolution mechanisms in real networks. Many similarity indices have been proposed based on a topology structure for link prediction. Local topological information, especially common neighbors, plays an important role in calculating the similarity between two endpoints. However, plenty of similarity indices ignore the effectiveness of common neighbors under different topology structures. Considering the local topological information around common neighbors, an effective common neighbor index is proposed. Firstly, we analyze the effectiveness of all neighbor links of common neighbors. Then, based on the local topology on both sides of two endpoints around common neighbors, the effectiveness of two sides of common neighbors is quantified separately. Finally, the similarity between two endpoints is described through the effect of common neighbors' effectiveness on bilateral resource allocation process. Empirical study on 15 real networks shows that the index proposed can achieve higher prediction accuracy, compared with 9 mainstream baselines.