Research On The Hierarchy Of Nodes And Controllability In Attention Flow Network

Influential nodes in complex network are critical to the flow and control of information,which maintain the cohesion of the entire network.Although there have been plenty of researches on the specification of influence of nodes and area of network controllability,there are few researches on the problem about the hierarchy of influence of nodes in directed and weighted network,and how to further control the network effectively.Directed and weighted attention flow network is a kind of network model that has emerged in recent years,whose research on the hierarchy of nodes and network controllability has important theoretical significance and application value.It not only can help analyze network centrality,node clustering and community characteristics etc.,but also help applications such as network regulation,accurate advertising and interest recommendation.However,there are few researches on the hierarchy and controllability in attention flow network.To solve this problem,the paper mainly does the following three aspects along the research idea of " The algorithm on influence of nodes —> The classification algorithm based on influence of nodes —> The controllability of nodes algorithm based on classification" :1.The algorithm for the hierarchy of influence of nodes in attention flow network is proposed.As for the problem with Hierarchical K-Shell for specifying influence of nodes traditionally in undirected and unweighted networks,by optimizing it,OHKS(Optimizing Hierarchical K-Shell)algorithm on the hierarchy of specification of influence of nodes that can apply to a directed and weighted collective attention flow network is proposed.First,constructing a directed and weighted collective attention flow network based on massive online users' online behavior log data provided by China Internet Network Information Center(CNNIC).Then,in order to specify the influence by defining two indices which are hierarchical position time(HPT)and position constraint(P)of nodes in the network,while considering time series and topological positions of nodes.Extensive experiments have found that OHKS algorithm can discriminate "illusive" nodes which seem to attract a large amount of attention but that aren't influential in fact,thereby can more accurately identify the most influential nodes.The results of research have important theoretical significance and application value in evaluating the influence of "star" nodes and network controllability.2.The algorithm for node classification in attention flow network based on GAT is proposed.Due to it is important that coordinate dynamic integration in a decentralized manner when controlling the network,the paper apply the efficient,flexible,and portable graph neural network model(GAT)to the learning of node classification in attention flow network,whose result is helpful for the study of classification controllability of network.First,selecting features of nodes and dividing categories of nodes by analyzing online users' online behavior log data,while constructing the attention flow network.Then,dividing nodes in the network according to practical application by using GAT model learning.Massive experiments show that although there are many non-linear associations and extremely complex regulations among data in the attention flow network,experimental results are relatively satisfactory.Comparing and analyzing the classification standards of authoritative Chinese websites worldwide,the result can meet the actual classification application requirements.3.The algorithm for classification controllability of attention flow network is proposed.By optimizing Structure-Driven MDS model for community controllability of complex network,CCN(Classification Controllability of Network)algorithm of attention flow network is proposed.First,constructing the attention flow network with distinct categories of nodes based on the algorithm and its result for nodes classification in attention flow network.Then,the problem about which controlling the entire attention flow network is transformed into controlling sub networks of various categories,by computing the adjacency matrix,ring structure matrix of various communities and using the branch and bound search strategy to determine their minimum dominating set.Extensive experiments have found that CCN algorithm not only determines controllable nodes based on topology structure,but also combines attributes of nodes while considering the professionalism and depth of vertical structure of nodes deeply,thereby control the entire network in a comprehensive and balanced manner.