Link Prediction Algorithm Based On Local Centrality Of Common Neighbor Nodes Using Multi-Attribute Ranking

Link prediction is intended to identify the missing links in the current network or predict the new links that will appear in the future based on some known information such as the node's attributes and network topology.Link prediction not only help us to reveal the inherent evolution mechanism of complex networks,but also help us solve many important practical problems.As a result,link prediction has become an important research topic of complex network.A large number of researches show some nodes may have greater influential in complex networks,and the similarity-based algorithms have low computational complexity.The existing similarity-based algorithms does not make full use of the important feature of node centrality,considering that the local centrality of common neighbor nodes have an important effect on the similarity-based algorithm,we proposed the link prediction algorithm based on local centrality of common neighbor nodes using multi-attribute ranking.The main work includes:Firstly,considering that some of the core nodes of the common neighbors have greater influential,especially in the social network,a large number of activities are around these core nodes.In this paper,the local centrality concept of common neighbor nodes and the improved local similarity-based indices are proposed.We also define the local closeness centrality,local betweenness centrality and local triangle centrality of the common neighbors,and use them as the local centrality measures in improved link prediction algorithm.Secondly,we proposed a multi-attribute ranking method based on the Technique for Order Preference by Similarity to Ideal Object(TOPSIS)to evaluate the local centrality of common neighbor nodes comprehensively,which avoid the problem that using the single index can not measure the local centrality of the common neighbor nodes accurately.Thirdly,consider the problem that the traditional coefficient of variation and entropy method does not apply to calculation the weight of the centrality decision-making indicators.In order to measure the importance of the centrality decision-making indicators accurately,we propose a weight calculation scheme based on the difference of decision metric,so the local centrality indicator based on TOPSIS can achieve better results.Finally,experimental studies on multiple real complex network datasets verified the superiority of the algorithm proposed in this paper.