Weak link prediction based on hyper latent distance in complex network

• Published in: Expert systems with applications Vol. 238; p. 121843
• Main Authors: Zhou, Ming-Yang, Wang, Fei, Chen, Ze, Wu, Ji, Liu, Gang, Liao, Hao
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 15.03.2024
• Subjects: Complex network; Link prediction; Weak link; Link prediction; Weak link; Complex network
• ISSN: 0957-4174; 1873-6793
• DOI: 10.1016/j.eswa.2023.121843

Weak links play a crucial role in the functionality and dynamics of networks. Nevertheless, the ability to forecast weak links accurately remains elusive. This article introduces a neural network framework that solely utilizes the network topology to predict weak links. Firstly, we embed the network into an embedding space and observe that weak links generally possess longer distances compared to strong links. Subsequently, we propose the concept of ’hyper latent distance’ as a means to characterize the pairwise node strength in the embedding space and incorporate it into our neural network model. Empirical investigations conducted on real networks demonstrate that our approach enhances the prediction accuracy for both weak and strong links concurrently. The potential of our method extends to the enhancement of recommender systems. •We explain why the traditional methods fail in predicting weak links.•We propose a topology-based method that aims to predict weak links.•We conduct abundant experiments to validate the effectiveness of our method.