Multilayer network link prediction considering multiple correlation features

• Published in: Expert systems with applications Vol. 285; p. 127700
• Main Authors: Zhao, Pu, You, Lan, Wang, Man, Peng, Qingxi
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.08.2025
• Subjects: Community detection; Complex networks; Layer similarity; Link prediction; Multi-attribute decision-making; Link prediction; Community detection; Layer similarity; Multi-attribute decision-making; Complex networks
• ISSN: 0957-4174
• DOI: 10.1016/j.eswa.2025.127700

Complex networks play a pivotal role in analyzing the multidimensional interactions within modern society. Link prediction, as an important downstream task, has demonstrated extensive application potential and significant research value in fields such as social networks, scientist collaboration networks, transportation and logistics, and energy and power grids. With the development of link prediction, research focus has gradually shifted towards multilayer networks that better represent actual interaction relationships. However, existing link prediction studies rarely focus on multilayer networks as primary research subjects. Even in studies targeting multilayer networks, the extraction and integration of multiple relational features remain limited, thereby constraining the performance and accuracy of link prediction. This paper aims to explore link prediction algorithms suitable for multilayer networks, proposing a method called Multi-Correlation Features based Link Prediction in Multiplex Networks (MCFMN-LP) that considers multiple relational features. By integrating interlayer, nodal, and community relational features, the study delves deeply into the impact of complex hierarchical and community structures on potential links between nodes in multilayer networks, aiming to enhance link prediction accuracy. It also introduces an interlayer similarity metric focused on the number of shared connections across layers and a community detection algorithm tailored for multilayer networks. Through comparative analysis of AUC and Precision evaluation metrics, the proposed method, which considers multiple relational features, is proven to have superior performance in multilayer network environments compared to single-feature methods. Moreover, compared to deep-learning-based link prediction models, MCFMN-LP achieves high predictive accuracy while maintaining lower computational costs, making it a scalable and efficient solution for large-scale multilayer networks. [Display omitted] •Proposed a multilayer network link prediction method integrating multiple features.•Introduced a multi-attribute decision framework to combine correlation features.•A common connection index measures structural similarity between network layers.•Likelihood estimation calculates community scores, capturing community association features.