Improving Node Classification Accuracy of GNN through Input and Output Intervention

Graph Neural Networks (GNNs) are a popular machine learning framework for solving various graph processing applications. This exploits both the topology and feature vectors of nodes. One important applications GNN is in semi-supervised node classification task. The accuracy using depends on (i) number (ii) choice training In this paper, we demonstrate that increasing nodes by selecting from same class spread out across non-contiguous subgraphs, can significantly improve accuracy. We accomplish presenting novel input intervention technique be used conjunction with different methods to increase and, thereby, also present an output identify misclassified relabel them their potentially correct labels. real world networks our proposed methods, individually collectively, comparison baseline algorithms. Both agnostic. Apart initial set generated techniques do not need any other extra knowledge about classes Thus, modular as pre-and post-processing steps many currently available