Graph-Assisted Bayesian Node Classifiers

Many datasets can be represented by attributed graphs on which classification methods may of interest. The problem node has attracted the attention scholars due to its wide range applications. consists predicting nodes’ labels based their intrinsic features, features neighboring nodes and graph structure. Graph Neural Networks (GNN) have been widely used tackle this task. Thanks structure they are able propagate information over aggregate it improve performance. Their performance is however sensitive topology, especially degree impurity, a measure proportion connected belonging different classes. Here, we propose new Graph-Assisted Bayesian (GAB) classifier, designed for classification. By using theorem, GAB takes into consideration impurity when classifying nodes. We show that proposed classifier less complex than GNN-based classifiers.