Research On Link Prediction Algorithm Based On Deep Convolutional Neural Network

Link prediction,as an important branch of network science,has been widely studied by many scholars in recent years.On the one hand,link prediction has scientific research value.For example,the current society is moving from the computer and Internet era to the Internet of Things.In the era of people,people,people,machines,machines and machines produced very complex relationships,which can be explored by establishing link prediction models;On the other hand,link prediction has strong practicability.For example,specific application scenarios such as friend recommendation in social networks,prediction of the stock market,and prediction of interactions between protein molecules.Therefore,link prediction in complex networks has very important research significance.Existing link prediction methods based on the network structure are mostly based on the common neighbors of nodes,and the common neighbor index of the nodes are usually obtained by extracting the simple neighborhood index information of the nodes.In addition,most of these methods use simple models that cannot extract valid link forms,resulting in poor link prediction results.Therefore,this paper starts from the network structure,focuses on the high-priced neighborhood index of nodes,and designs an effective extraction method,combined with the deep convolutional neural network model in the field of deep learning to conduct research.The main work of this article is as follows:(1)A link prediction model based on node neighborhood subgraph(NNS-LP)is proposed.The model is based on a convolutional neural network,extracts the high-order neighborhood network feature information of the nodes through a subgraph extraction algorithm,and organizes this information into a feature information matrix.Finally,NNS-LP is used to determine whether there is a link between nodes.Fully conducted experiments on the public data sets of USAir,King James,Metabolic,and PB,and compared with the current link prediction algorithms based on network representation learning,both the AUC index and the Precision index have a certain degree of improvement.To a certain extent,NNS-LP effectively overcomes this type of algorithm focusing on extracting feature information from the low order neighborhood of nodes.At the same time,the model can adapt to network data in different fields and can greatly improve the effect of link prediction.(2)A link prediction algorithm model based on multi-order adjacencies and clustering coefficients is proposed(MLANCC).By introducing the attribute of the node's clustering coefficient and combining the two important features of the node's multi-order adjacency nodes,the clustering coefficient of the node is calculated,and different weights are given according to the proximity of the node,which is used to evaluate the nodes.There is a possibility of link.According to the experimental simulation results on the public experimental data sets USAir,NS,King James,metabolic,PB,and UA,The MLANCC algorithm on AUC and Precision index is mostly better than the benchmark algorithm.It shows that MLANCC can capture more link feature information than existing algorithms based on common neighbors of nodes,and can deal with most small and medium-sized networks,so the algorithm has higher classification accuracy and prediction effect.