Graph Matching Algorithm Based On Attention Graph Convolution Network And Deep Belief Network

With the rapid development of Internet technology,more and more graph data information accumulates rapidly.Graph matching,as a basic algorithm for mining valuable information contained in graph data,has become a research hotspot in this field.As deep learning is applied to solve graph matching problems and shows a development trend that is superior to traditional methods,graph matching algorithms based on deep learning theory have developed rapidly in the past two years.Among them,the convolutional neural network has received extensive attention from researchers due to its efficient performance and fast learning ability.The current deep learning model has too many parameters and a large amount of calculation,which leads to high algorithm complexity and insufficient data mining.In response to the above problems,this article has carried out the following work:(1)When the existing graph convolutional network model is applied to inaccurate graph matching,the node characteristics and topological features between nodes are easy to lose in the early convolution step,and an improved attention graph convolutional network model is proposed..Use relatively few parameters to learn hierarchical representation in an end-to-end manner,and use a self-attention mechanism to distinguish nodes that should be discarded or retained.First,use the attention graph convolutional network to automatically learn the importance of different jumps on the neighborhood;second,add a self-attention pooling layer to summarize the graph representation from all aspects of the matrix graph embedding;and then send the final graph embedding matrix M to the linear layer to Make the final prediction;finally,train and test in multiple standard graph data sets.Experimental results show that compared with the most advanced graph kernels and other deep learning algorithms,this method has better graph classification performance on standard graph data sets.(2)In the traditional deep belief network,because the graph data is distributed anywhere in the neighborhood,it is easy to cause the loss of important information such as topological features in the graph.In order to solve this problem,a graph matching algorithm combining deep belief network and convolutional neural network is proposed.First,learn the local structural features of the data through CNN,and then introduce the convolution operation in the DBN model to establish a convolutional deep belief network composed of two CRBN stacks,and finally adopt a greed similar to DBN in multiple standard graph data sets.The layer-by-layer training method and the Softmax classifier train and test the convolutional deep belief network.Experimental results show that this method is superior to existing methods in image classification and retrieval.The method proposed in this paper can effectively solve the problem of insufficient node feature and topological feature mining,and the constructed model can effectively improve the accuracy of graph classification and the efficiency of graph retrieval.