Link prediction based on node weighting in complex networks
- PDF / 1,554,159 Bytes
- 16 Pages / 595.276 x 790.866 pts Page_size
- 56 Downloads / 219 Views
(0123456789().,-volV)(0123456789(). ,- volV)
METHODOLOGIES AND APPLICATION
Link prediction based on node weighting in complex networks Og˘uz Fındık1 • Emrah O¨zkaynak1
Springer-Verlag GmbH Germany, part of Springer Nature 2020
Abstract Link prediction is used to predict future links in complex networks. Traditional methods proposed for link prediction make estimates based on similarity measurements, taking into account only the instant topological structure of the network. However, especially in dynamic networks, the activity of nodes varies over time, so it is not enough to measure similarity from topological properties for a good prediction process. Accordingly, the success rate is low in prediction processes where the power of the nodes in the network is not sufficiently reflected. In this study, a novel link prediction model called ‘‘Link Prediction Based on Node Weighting in Complex Networks’’ is proposed to overcome the mentioned problems. Unlike using weights between nodes, the proposed model is based on calculating the own weights of the nodes and making the link prediction. The weighting process includes factors such as eigenvector centrality, experience, continuity that can reveal the power of nodes over time. The model consists of two parts. The first part is node weighting, which calculates the strength of nodes in the network. The second part is the node-weighted link prediction process, where node weights are used to predict future links. Scientific collaboration data at IEEE Xplore and Australian Open Tennis Tournaments data were used to test the success of the proposed model. In experimental studies conducted in networks created from different time periods, it has been determined that the proposed method gives more successful results than the latest technology methods according to the AUC metric. Keywords Complex networks Link prediction Node-weighted networks Social networks Multi-criteria decision analysis (MCDA)
1 Introduction Networks are widely used to represent systems created by every entity that communicates or interacts with each other ¨ zkaynak 2018). Developments in information (Fındık and O and communication technology make important contributions to the creation and growth of networks. With the emergence of rapidly expanding networks, the analysis of the entities that forming the networks and the relationships between them has gained importance (Subbaraj and Sundan 2018). Complex network science provides the ability to analyze any structure and system within the framework of
Communicated by V. Loia. ¨ zkaynak & Emrah O [email protected] Og˘uz Fındık [email protected] 1
certain rules and disciplines that have direct or indirect relationships between them (Strogatz 2001). Link prediction in networks involves revealing missing links based on the information obtained from the attributes of the observed links and nodes and the structural features of the network or predicting new links that may occur in the future (Liben-Nowell and Kleinberg 2007; Linyuan and Z
Data Loading...
