Improvement And Application Of Correlation Measurement Of Heterogeneous Networks Based On Probability Graph

Most data or information objects in the real world are interconnected or interact with each other,forming numerous large,interconnected,and complex networks,which are called heterogeneous information networks.It can be used to describe actual networks,including social relations between people and cooperative relationships in scientific research.Research on complex networks helps to discover the implicit knowledge in the networks,which has become a common focus in many disciplines.Of these,the most basic and important problem is the correlation metric.This paper investigates node correlation metrics and relationship prediction based on heterogeneous network paths.By constructing a path-based probability graph model of heterogeneous networks,taking the correlation metrics in heterogeneous networks as the main research tool,and combining the dynamic features in heterogeneous networks,we focus on the probabilistic relationship between node pairs with different attributes.Subsequently,in combination with the Matrix Factorization methods,the goal of effective recommendation for each user in a dynamically changing heterogeneous network is achieved.The main contributions and work of the thesis is as follows:First,a correlation measurement called DHIN-PReP based on the meta path is proposed.The Probabilistic Graphical Models are used to construct the node relationship probability function.A priori information is then added to the path-selective feature of the heterogeneous network by assigning weights;Subsequently,time information in the network is extracted and an effective time-influence function is designed.And the correlation metric function and the time impact function are combined to obtain a new path-based dynamic correlation metric,abbreviated as DHIN-PReP.Second,based on the correlation score matrix obtained by the DHIN-PReP model,a correlation recommendation algorithm based on matrix factorization is proposed in this paper.From the perspective of probability,both PReP and DHIN-PReP correlation measures can represent complex information network relationships.Therefore,in order to improve the effectiveness of recommendation on complex heterogeneous information networks,a new recommendation algorithm based on probability correlation score matrix is designed,abbreviated as PReP-MF and DHINPReP-MF.The algorithm builds a user relationship model for network nodes,and obtains a correlation score matrix,mines the relationship information in the network,and then decomposes the matrix to obtain the relationship prediction results between pairs of nodes with missing correlation scores,sets reasonable thresholds,and finally to achieve recommendations according to user needs.Finally,validation experiments on the relevance metrics proposed in this paper are performed on the datasets based on Facebook and DBLP.The experimental results show that the effectiveness and stability of the DHIN-PReP algorithm is better than the existing algorithms.While ensuring the accuracy of the node’s judgment of the relationship,it also improves the reliability of the relationship prediction and reflects the dynamic change process of the relationship in the heterogeneous networks.At the same time,the correlation recommendation algorithms PReP-MF and DHINPReP-MF were experimented using the correlation scoring matrix obtained from two real data sets,DBLP and Facebook,and the results proved that the new algorithm performed well in several performance evaluations,and the matrix incorporating the time information was better than the matrix recommendation obtained from the PReP algorithm.