Path-based estimation for link prediction

Guoshuai Ma, Hongren Yan, Yuhua Qian, Lingfeng Wang, Chuangyin Dang, Zhongying Zhao

Abstract:

Link prediction has received a great deal of attention from researchers. Most of the existing researches are based on the network topology but ignore the importance of its preference; for aggregating multiple pieces of information, they normally sum
up them directly. In this paper, a path-based probabilistic model is proposed to estimate the potential connectivity between
any two nodes. It takes carefully the effective influence of nodes and the dependency among paths between two fixed nodes
into account. Furthermore, we formulate the connectivity of two inner-community nodes and that of two inter-community
nodes. The qualitative analysis shows that the links between inner-community nodes are more likely to be predicted by
the proposed model. The performance is verified on both the multi-barbell network and Lesmis network. Considering the
proposed model’s practicability, we develop an algorithm that iterates over the adjacent matrix to simulate paths of different lengths, with the parameters automatically grid-searched. The results of the experiments show that the proposed model
outperforms competitive methods.