The Research Of Link Prediction In The Bank Transaction Network

Link prediction,as an important research direction of complex network,solved the fundamental problem of complex networks--reduction and prediction of missing information.In short,link prediction is to predict the probability of edge-to-edge connection among nodes in the network that has not yet been connected by known nodes in the network.At present,there are two main types of link prediction algorithms in the mainstream.One is based on Markov chain or machine learning.The algorithm mainly considered the attributes of nodes in the network.It can achieve better prediction accuracy in the aspect of prediction effect but is limited a lot of constraints on the parameters.The other type of algorithm is based on the maximum likelihood estimation of the network structure.Such algorithms have high computational complexity in general conditions.It is not suitable for complex networks with large data volumes.It can be seen that the traditional link prediction is mainly focus on undirected and unweighted networks.It has less research on undirected and unweighted networks,directed and unweighted networks,directed and unweighted networks.In this thesis,the research object of this thesis is the bank transaction network.Based on the theory and technology of link prediction,the learning model is introduced.The structural attributes and evolution trend of the network are considered comprehensively and connected nodes of self-adaption algorithm(CNSA)is designed and implemented.The main research contents of this thesis are as follows:First of all,the thesis introduces some basic conceptions and unique characteristics in social network analysis and link prediction and the research development of link prediction combined with the analysis of existing link prediction algorithms and evaluation indicators based on different characteristics.Secondly,through studying on transaction data of bank customers,we construct an undirected and weighted model that is applied to the customer transaction network,and then,the client trade network topology quantitatively is analyzed,such as the node degree,degree distribution,average path length,clustering coefficient.It is verified that the customer transaction network have the small-world and scale-free characteristics of complex networks.Thirdly,due to the dynamic characteristics of trading networks,a dynamic prediction algorithm,which can be suitable for trading network are developed.Through the change of the weight of nodes,the weight of nodes is adjusted dynamically so that it can satisfied the dynamic growth characteristics of the network.The network is predicted based on the results of the training set.After that,the algorithm is improved by combining the strength of network nodes and the node centrality.Finally,based on the actual data set experiment,the prediction accuracy of the algorithms and the existing classical link prediction algorithm are compared and analyzed.It can be seen that the prediction accuracy of the proposed algorithm is better than the prediction accuracy of the existing algorithms.