Network-energy-based predictability and link-corrected prediction in complex networks

Abstract

The existing link predictability indexes for networks often depend on specific structural characteristics and don’t make full use of all the structural information of networks. This paper finds that network energy can efficiently characterize the information of the network structure. Firstly, a novel and simple network predictability index is proposed via normalized network energy. To enhance the global topology information of the predictability index, we then extend this index and define a new effective predictability index by integrating the network structural consistency index. Secondly, based on modified maximum likelihood probability method, we develop the approximate relationship between the probability of the original network (subject to the condition of the observation network) and the maximum likelihood probability of the perturbed network. Furthermore, a novel link prediction algorithm (LCPA) is presented. Numerical experiments on both generative and real networks confirm that the LCPA algorithm outperforms the existing state-of-the-art methods in most cases. Finally, this paper takes the precision obtained by the LCPA algorithm as the network predictability value and also verifies the effectiveness of the two proposed predictability indexes. It is also shown that the above-defined indexes can efficiently characterize the network predictability. The latter defined index also shows a linear relationship with the network predictability. The data and code are publicly available at https://github.com/pinglanchu/LCPA.