Research On Coupled Network Link Prediction Algorithm

Link prediction,aims to predict the emergence and disappearance of links in the network evolution,is a core problem in social network data mining.However,most of exiting work focuses only on single network,and there is a little research on link prediction of coupled network.Many real-world networks interact with and depend on each other,so we have to take into account the interaction among networks instead of only single network.Coupled network link prediction had problems of low generalization of the meta-path and neglecting the influence of different meta-paths.Therefore,it is necessary and meaningful to study problem of coupled network link prediction.We integrated the unified meta-path scheme and implemented coupled network link prediction algorithm(CNLP)to solve problems.Experiments showed that the proposed algorithm is close to the original meta-path method,but it can improve generalization of algorithm,treat differently different meta-paths and not need that experts designed meta-paths.We applied link prediction theory to link prediction analysis system,which facilitated intuitive visualization of link analysis effect and process through visualization.We completed potential white-list and product recommendation through specific cases,the main work of paper includes two aspects as follows:(1)We improved similarity index(CN,Katz)and coupled network link prediction algorithm.We integrated node guidance-force and time factor in the similarity indicator.The node guidance-force considered different roles of nodes and paths in the network;the time factor measured the probability of forming connection by average degree.This method is also effective in other similarity and path indicators.Experiments showed that experiment accuracy is only increased by nearly 2%compared to CN and Katz indicators when using node guidance-force,at the same time,we used node guidance-force and time factor method,which is improved by 5%to 10%.This paper constructed the unified meta-path method to solve heterogeneity of coupled network,treated differently types of meta-paths in the network and integrated coupled network link prediction algorithm(CNLP).Experiments show that this method improved accuracy over other methods by more than 3%.the unified meta-path enhanced generalization of model,treated different meta-paths in the network and designed automatically meta-paths.(2)We designed and implemented link prediction analysis system about complex network,which is used to collect information for completing online review,viewing related data,analyzing and managing links.The system adopted the SOA distributed architecture design principles deepening interface design.We focus on implementing core modules of system,rationalizing architecture design and providing exhaustive solution scheme of system.The implementation of system showed that the distributed architecture facilitated to integrate system resources,improve stability,flexibility and scalability of system,provide a good interface and enhance the stability and security of link analysis,so it had a wide range of practical values.The system integrated link prediction theory with internet finance to help discovering potential customers and recommending financial products.