Research On Dynamic Link Prediction Based On Triple Motif Evolution

In the network,the task of Link Prediction aims to learn the potential relationship between nodes to predict the unknown potential connection state.Most of the current Link Prediction methods are used to deal with static networks,but in reality,most of the networks belong to Dynamic Networks,that is,the vertices and links of the network will change with time;therefore,these methods cannot produce good prediction results for Dynamic Networks.In order to improve the accuracy of Dynamic Network Link Prediction,this paper starts from the perspective of network microstructure evolution,on the basis of Dynamic Network Time Window partition and optimization,introduces the integrated moving average autoregressive model to build the Probability Matrix of prediction of Motif Evolution.Considering the influence factors of Motif Evolution and Motif Evolution probability,the connection edge probability between any nodes can be obtained.The main contents of this paper are as follows:Research on time window partition method of Dynamic Network.In order to quickly and accurately partition the appropriate size of Time Window under the condition of low loss of network information,this paper determines the function that takes Time Window as the common variable and presents the opposite trend in two Dynamic Networks,and uses the difference between the two functions to minimize to find the appropriate size of window.In this paper,we propose a convenient way to deal with the dynamic of the network,which will greatly help to improve the overall efficiency of the algorithm.Build time series prediction model.According to the characteristics of Dynamic Network,an Integrated Moving Average Autoregressive Model is constructed,and the time series obtained in the second step is regarded as random series.The correlation of these random variables reflects the continuity of the original data in time,so as to build a prediction matrix of the probability of Motif Evolution according to the model.At the same time,considering the formation of the connection edge of the link weight and closure triple in the process of Motif Evolution Influence,synthesize the prediction matrix to get the probability of any two nodes to produce the connected edge.In this paper,two real data sets Enron network and Facebook wosn wall are tested.Firstly,the validity of the time window partition method is verified,and the AUC prediction index is used as the evaluation standard to compare the MFME method with TTM,TCM and TS.The results show that the proposed method can achieve better link prediction results.