Predicting missing links in complex networks via an extended local naïve Bayes model

¹ Key Laboratory of Intelligent Computing and Signal Processing of Ministry of Education, School of Mathematical Science, Anhui University - Hefei 230601, China
² School of Internet, Anhui University - Hefei 230601, China

^(a) chuang_m@126.com

Received: 6 November 2019
Accepted: 25 May 2020

Abstract

In this work, we propose an extended local naïve Bayes (ELNB) model to implement link prediction in complex networks. In the model, an extended clustering coefficient (ECC) is defined to represent the posterior connection probability in ELNB model, which is composed of the link clustering coefficient (LCC) and the node clustering coefficient (NCC). Our method not only overcomes the shortcoming of the common neighbor similarity index —each common neighbor of two nodes contributes equally to the connection likelihood— but also guarantees that the contributions of a common neighbor on the connection likelihood of different pair nodes in its neighbor set are different. Through the in-depth analysis, we find that the LCC plays a positive role on link prediction. On the contrary, the NCC often yields a negative effect. Therefore, we can find the optimal parameter in ELNB yielding the best link prediction. More importantly, we prove that ECC is different from clustering coefficient in the micro-scale (individual level), but they are the same in the meso-scale (subnetwork level) and macro-scale (whole network level). As a result, ECC is a new and effective index to characterize the micro-structure of networks while it retains the invariance property in meso-scale and macro-scale.