Research On Multi-order Pattern Analysis Based On Network Representation Learning

In the current era of rapid development of network information,a large amount of network data has been generated.How to obtain valuable information from massive data network is cutting-edge research in the field of current researchers concerned.A large number of vertices is distributed in the network.Because of different attributes and types,complex interlaced connections are formed.These connections imply rich feature information.The network representation learning algorithm is specifically used to represent feature information in the network.Provide input to existing machine learning algorithms by encoding network vertices into the form of a vector of real numbers.The network topology in vector space is closer to the real structure and can be applied to research work such as vertex classification,visualization,community discovery,and link prediction.This paper first introduces the concept of network representation learning and the development of some existing algorithms.A new network representation learning algorithm BD-LargeVis is proposed,which is an improved optimization of the LargeVis algorithm,mainly for complex network visualization problems.The concept of backbone is incorporated into the algorithm to improve the similarity of adjacent vertices in the network and apply it to visualization and community discovery.The final classification results were obtained using k-means clustering in community discovery experiments,and the experimental results were evaluated using the F-measure evaluation algorithm.Experiments show that the proposed algorithm compares with the first-order mode and the second-order mode of LINE,and the F-measure value is higher.In order to study the application of network representation learning in link prediction,a multi-order mode algorithm BDLINE based on backbone and network coding is proposed.The algorithm integrates the backbone algorithm based on the first-order similarity and the second-order similarity of the network representation learning algorithm LINE,and encoding the network into a multi-dimensional vector space,making the probability that the vertices corresponding to the highly similar vector appear edges is greater.The vertices in the network structure are represented by vectors,which is more convenient for extracting network feature information,and can better deal with the randomness problem of social network in link prediction.The experiment used two different real data sets and used different algorithms to perform multiple experiments.In the experimental aspect of link prediction,compared with other network representation learning algorithms,the performance of the algorithm model BDLINE proposed in this paper is improved,and the prediction effect is better.It has certain guiding significance for predicting the relevance and effectiveness of network vertices in complex social networks.