Research On Social Network Relationship Based On Functional Network

At present,social networks have led to fundamental changes in the way users interact.In order to reveal the potential relationships between users in social networks,the prediction of social relationships has become a basic problem in the study of social networks.The prediction research of social relationship is mainly based on the calculation of similarity between users,which is divided into traditional methods and deep learning methods.Traditional methods have strong dependence on the structural information of social networks and low experimental accuracy,while the methods based on neural networks(NNs)have high accuracy.However,the structure of neural network is a "black box".Now,many related researches on interpretable models becomes more and more important while guaranting the accuracy rate.In order to address the above problems,this thesis proposes a new Functional Network(FN)model,which constructs the connection between the neurons of the network through the related association rule algorithms.The main content is as follows:1.We analyze the user attribute characteristics of social networks and mines its influence on the relationship prediction results.Firstly,the cloud model is used to divide each attribute value into two clusters to obtain the degree of certainty of the attribute;Secondly,the above two-classification result is used to calculate the frequent two itemsets for the related attributes,and a two-layer FN Model structure is constructed.This method overcomes the high dependence of social networks structure and makes the model be interpretable also.2.In order to further improve the prediction effect of the model,combined with three-way decisions(3WD-FN),an improved FN model is proposed.Firstly,the three-way decisions theory is used to divide the prediction results output by the two-layer FN model.The prediction results outside the prediction boundary can be directly output,but the uncertain prediction results within the preset boundary will be input into the third-layer of our FN model to predict further.3.Because of the problem that the parameters cannot be solved in the FN for some special cases,a functional network model based on gray correlation analysis(GRA-FN)is introduced.In this model,the correlation between the attributes can be obtained without the discretization.Because the Apriori algorithm relies on the degree of support to obtain the degree of association between attributes,different support degrees can get different degrees of association,and the degrees of association between certain attributes will be ignored easily.However,gray correlation analysis does not abandon every possible attribute combination,that is,every possible combination can get the corresponding correlation value.The model only builds a twolayer network structure,and the active function of the output layer neuron is a multivariate primary function.It avoids the problem that parameters cannot be solved.