Research On Feature Selection Algorithm Of Graph Neural Network

With the rapid development of the Internet,the scale of data in real life has become increasingly large.How to extract the data we are interested in from massive amounts of data has become an important issue.With the development of deep learning,people have become more and more proficient in processing Euclidean structured data.However,there are also non-Euclidean structured data in real life besides Euclidean structured data,namely graph-structured data,which is very common and contains a huge amount of information.Common deep learning models perform poorly when dealing with graph-structured data.Based on this,graph neural networks were proposed for processing graph-structured data and they have strong interpretability that is convincing.With the development of graph neural networks,their family is becoming larger and their applications are becoming wider.Now,Graph Neural Network(GNN)have become a useful tool for solving a variety of practical problems.In recent years,researchers have proposed many improved models,driving performance improvements in various prediction tasks.But most GNN variants start with the assumption that node features are equally important,which is not the case.In response to this issue,this article proposes a new feature selection algorithm to improve existing models and select important feature subset for the dataset.The algorithm consists of two components,a feature selection layer and a separate label-feature mapping.Using softmax normalizer and feature "Soft Selector" for feature selection in the feature selection layer,and the model structure is designed under the idea of separate label-feature mapping to select the corresponding subset of related features for different labels,and multiple related feature subsets are combined and calculated to obtain an important feature subset of the final dataset.The Graph Attention Network(GAT)and GATv2 models are selected as the benchmark model,and the algorithm is applied to the benchmark model to obtain a new model.Experiments show that when the new model performs node classification tasks on six data sets,its accuracy rate is improved by 0.83%to 8.79%compared with the baseline model;the new model also selects the corresponding important feature subsets for the six datasets,their number of features accounts for 3.94%-12.86%of the total number of features in their respective datasets.After using the important feature subset as the new input of the benchmark model,the accuracy rate of more than 95%of the benchmark test results(using all features)is still obtained.That is,the scale of the model is reduced on the basis of ensuring the acicuracy.It can be seen that the proposed new algorithm can improve the accuracy of node classification,and can effectively select the corresponding important feature subset for the dataset.