Active Learning for Semi Supervised Node Classification with Selective Features

Abstract

Node classification is a fundamental process for graph information which basically establishes node classifications based on their properties and structure of graph. Current learning algorithms from graphically represented information have increased interest in Graph Neural Networks (GNNs) for predictive problems including node categorization or edge forecasting. The genuine success of GNNs is considerably constrained by the difficulty of obtaining a large number of node labels. In semi-supervised classification, active learning entails adding new labels to previously unlabeled data to boost the underlying classifier's performance. This paper presents active learning for semi supervised node classification (AL-SSC) with selective features. This approach use feature learning methods for reduce the dimensionality of the data and use SOM cluster-based active learning model for node categorization that utilize labeled and unlabeled data. Extensive experiments were used to analyze the accuracy of the suggested work's performance. Experiments are conducted using the three publicly accessible reference datasets Citeseer, Pubmed, and Cora. According to the findings, the suggested technique performs well compared with other previous methods.