We describe a method to select the nodes in Graph datasets for training so that the model trained on the points selected will be be better than the ones if we select other points for the purpose of training. This is a very important aspect as the process of labelling the points is often a costly affair. The usual Active Learning methods are good but the penalty involved with these methods is that, we need to re-train the model after selecting the nodes in each iteration of Active Learning cycle. We come up with a method which use the concept of Graph Centrality to select the nodes for labeling and training initially and the training is needed to perform only once. We have tested this idea on three graph datasets - Cora, Citeseer and Pubmed- and the results are really encouraging.
View on arXiv