Research On The Similarity Of Nodes In Attention Flow Network

Attention flow network is an important branch in the field of the research about complex network.The similarity of nodes is a key method to characterize the structural characteristics and nodes attributes of the complex networks.The research on the similarity of nodes in attention flow network not only helps to unfold theoretical research work such as network community detecting and link prediction,but also provides reliable methods to measure the website influence,website classification and ranking,has an important theoretical and practical value.Although the application of the similarity of nodes in attention flow network is very extensive,there are few researches on it.Two similarity algorithms are proposed in this paper to measure the similarity of nodes in attention flow network,our research contents as follows:Firstly,based on the log data of user’s online click behavior provided by China Internet Information Center,the main website of user’s online click in the log data were extracted to generate the main website list.And then the attention flow network is constructed by using network science,graph theory and main website list.Secondly,based on the optimization of SPA(spatial preferred attachment)model,NID algorithm(Nodes Influence Distance algorithm)is proposed to measure the similarity of nodes in attention flow network.In this algorithm,SPA model is optimized by defining node generation time(_tR)and node influence radius(S_r)of the attention flow network;then a new node quantization form V_s(7)R _t,S_r(8)is defined.secondly,influence distance(S_d)between nodes is calculated by using space 2 norm based on the V_s(7)R _t,S_r(8).finally,S_d is used to quantify node similarity.Experiments show that:the R_t distribution andS_r distribution are heavy-tailed distribution;whenR_t is early andS_r is large,theS_d is small and the similarity between them is higher;the similarity between nodes of the same type is higher.Thirdly,the RE-NSM algorithm(Relative Entropy-Nodes Similarity Matrix algorithm)is proposed to measure the structure similarity of node in attention flow network based on the theory of relative entropy.There are four steps in re-nsm algorithm.Step one:dividing the attention flow network and generating several local networks.Step two:counting the structural characteristics of nodes based on the local networks.Step third:transforming the structural characteristics of nodes into node structural information,and construct node structural information probability set from node structural information;Step fourth:Quantifying the structural similarity between each pair of nodes by calculating the relative entropy of node structural information probability set.The research shows that:Large scale nodes have the characteristics of strong attraction,long survival time,high probability of users’priority click;The structural similarity between large-scale nodes is high,and some nodes with low similarity to large-scale nodes have died,which shows that the attraction of such nodes is weak,unique(only clicked once by the user),short survival time,low probability of being clicked by the user,and some nodes are only clicked by one or a few people.