Subgraph feature extraction based on multi-view dictionary learning for graph classification

Subgraph feature extraction of graph data has an efficiency problem that has become increasingly significant. A new architecture of subgraph feature extraction named GMADL is proposed in this paper. Dictionary learning approaches are put forward to extract the features of graph data to enhance the discrimination of model. To improve the efficiency of extraction, the analysis dictionary is designed as a bridge to generate the sparse code directly. Each sparse code represents the feature matrix of a graph. Through constructing the multi-view support vector machine (SVM) classifiers, the problem can be transferred into the multi-view problem so that the information of the whole view is utilized to predict the classification model. The comparison of the proposed architecture with the state-of-the-art approaches manifests the feasibility and the competitive performance in graph classification. (C) 2020 Elsevier B.V. All rights reserved.

Automated summary

The paper proposes a new architecture named GMADL for subgraph feature extraction, utilizing dictionary learning approaches to enhance discrimination of model features in graph data. By designing an analysis dictionary and constructing multi-view support vector machine classifiers, the efficiency of feature extraction is improved and the classification model prediction accuracy is enhanced by utilizing information from multiple views. Comparisons with state-of-the-art approaches demonstrate the feasibility and competitiveness of the proposed architecture in graph classification.