Research On Graph Matching Algorithm Based On Deep Learning

In recent years,with the explosive growth of information data and the extensive application and research of deep learning methods in various fields,the Graph structure is used as a basic structure representing the relationship between data,and it passes through a set of objects(nodes)and relationships(Modeling,often used to express largescale and complex data.For example,in practical applications,how to query a highly relevant match in this particular network relationship can be formalized as a matching problem of the graph to solve.The current graph matching methods mainly include two kinds of accurate map matching and inaccurate graph matching.Due to external deformation and noise,different degrees of difference may exist between different graphs from the same pattern,so the difference between graph structure and label is considered.Inaccurate graph matching technology has gradually become a research hotspot.With the increasing research on deep learning by experts and scholars,a large number of methods based on deep neural networks are applied to the graph matching problem,while the existing algorithms for deep learning processing graph matching only propagate information through the edges of the graph.You cannot infer and aggregate information in a hierarchical way.Although good results were obtained on the graph matching task,there is insufficient characterization of the planarization and graph data.In response to the above questions,the research content of this paper is as follows:(1)Drawing on the relationship theory of space-space syntax,the topological feature extraction on the original sample of the graph data is performed,and then the extracted feature data is preprocessed as the input of the deep learning network.(2)In-depth study of the working principle of the graph convolutional neural network,combining the spatial syntax and the graph convolutional neural network,and defining the feature information and structure of the graph nodes by defining convolution and micropooling operations on the graph.The information is layered end-to-end learning,which can more fully and efficiently represent the graph.Finally,the high-dimensional feature map is mapped to the low-dimensional topological space through the classifier to complete the classification task.Finally,the experiment proves that the model solves the problem of characterization of graph data and improves the accuracy of matching to some extent.(3)In-depth study of the cyclic neural network and LSTM neural network,this paper uses the idea of LSTM neural network to construct a graph matching algorithm based on LSTM neural network,which can model the structured data and mine the connection inside the graph.The relationship forms an overall understanding of each graph sample,and finally the classification task is completed.Experiments show that the method effectively improves the accuracy of classification.In summary,this paper sorts out the shortcomings of the existing algorithms and determines the improvement direction of the current algorithm.Firstly,the input of the deep learning model is established by the spatial syntactic variables.Secondly,two deep learning models are constructed,which solves the problem of insufficient representation of the current graph data,and then the GPU environment is configured and used in the model training process.Finally,the feasibility of the proposed algorithm is proved by experiments,and the accuracy of classification can be effectively improved.