Link Prediction Based on Multi-steps Resource Allocation

Abstract

Link prediction is a fundamental problem in social network analysis. The key technique in link prediction is to find an appropriate similarity measure between nodes of a network. Generally, external information besides the network topology is considered in many similarity measures. However, these external information is generally not available or not true. Usage of these external information may result an improper suggestion. Can we just use the information of the network topology to predict the miss links in the network? In this paper, firstly, we discussed the performance of 10 similarity indices, which only use network topology, on seven real networks. Secondly, for improving Resource Allocation index, which only used the common neighbors to calculate the resource of destination nodes gotten from source node, we proposed a new similarity measure called Multi-Steps Resource Allocation (MSRA). In MSRA, we used the information of multi-steps neighbors to transmit the resource from one node to another node. 2 steps neighbors, 3 steps neighbors and 4 steps neighbors are considered in this paper to be a balance in the performance and the computational complexity. Finally, to exhibit the power of MSRA in link prediction, we compare 10 various link prediction methods over 7 networks. The results show that our newly proposed MSRA measures outperform those ten measures on most datasets.