Hybrid Influence Index Based on Endpoint Attribute Diversity in Link Prediction in Complex Networks

Abstract

Prediction of links in complex networks aims to study possibility of forming a link between unconnected nodes, with the information contained in network, mainly including nodes, links and topological structure. In recent years, it has caused amounts of attention in various fields and achieved great success, especially in the indices based on topological similarity. However, most topological similarity indices incline to predict links based on transmission paths rather than endpoints. In addition, they typically measure endpoint influence only by degree, ignoring the role of other attributes of endpoints. Therefore, considering the diversity between endpoint attributes, we propose a method named Hybrid Influence Index based on Endpoint Attribute Diversity (HED), to specifically locate the impact of endpoint attributes in the measurement of endpoint influence. Finally, experiments based on twelve datasets of real world account for the superiority of our model in accuracy