Research On Link Prediction For Opportunity Network Based On Deep Learning

Link prediction is an important task in network analysis.Different from link prediction in static networks,in the opportunistic network,due to the mobility of the network nodes,the sparseness of the distribution,and the intermittent nature of the connection,there is no complete path between the source node and the destination node.Make routing forwarding a huge challenge.Link prediction in the opportunistic network refers to predicting the formation of future links through known network information,thereby reducing the blindness of message forwarding and improving the success rate of message forwarding.The existing link prediction algorithm has a good effect in static networks,but the opportunistic network is a typical dynamic network,so the existing link prediction algorithm is limited in the opportunistic network.The two main problems studied in this paper are: 1.Construct a link prediction algorithm that matches the opportunistic network.2.The law that the excavator will change the network link with time.The detailed contents of this dissertation are summarized as follows.1.Research on the similarity index of opportunity network.In order to make full use of network topology information and link history information,this paper divides the opportunistic network into several time slices,establishes a network time series,and then calculates the similarity index value of each link in the time slice.In this paper,considering the characteristics of opportunity network sparsity and intermittent connectivity,considering the spatial and temporal similarity of historical links,the concept of separation duration is proposed,and the average separation duration of similarity indicators that can reflect the timing of network links is constructed.2.Establishment of a time series model.Aiming at the dynamic characteristics of the opportunistic network,the Long Short-Term Memory(LSTM)model is constructed to transform the link prediction problem into a two-category problem.The model uses the similarity index proposed in the first step as the basic sample.Through the deep learning technology,the dynamic characteristics of the opportunistic network link are extracted,the time slice and the sample dimension are optimized,and the model is tuned.Then the model is used totrain the model.After the model training is completed,the model is used by the softmax classifier.The output is categorized to predict the links in the opportunistic network.In this paper,the experiment is conducted under the opportunistic network dataset Infocom05 and MIT.In the verification stage of similarity index,AUC and precision are used as evaluation criteria.The average separation duration indicator(ASLP)proposed in this paper is compared with the traditional similarity index to verify its effectiveness.Using the AUC,precision,and accuracy criteria of the Receiver Operating Characteristic Curve(ROC),the LSTM model proposed in this paper is compared with Naive Bayesian,Logistic Regression,and K-nearest neighbors.Common classification models such as(KNN)and Support Vector Machine(SVM)are compared.The test results show that the LSTM prediction model has better accuracy and stability.