Improving graph prototypical network using active learning

Abstract Due to the growth of using various devices and applications in modern life, amount data available is skyrocketing, but labeling all this beyond reach scientists. Thus, it necessary categorize with a small labeled data. In fact, should be possible prioritize for labeling. To achieve goal study, we have used few-shot learning active also power graph convolutional networks classifying graphical structure. implement proposed model, use two parallel calculate embedding importance each node. Using output both networks, create prototypes classes, then, classify them according distance node these prototypes. We select more intelligently, which improves overall model performance. As well as this, tested our field electronic commerce tagging goods big online stores, encounter large number diverse products, where high accuracy categorization short time without interference human factor help artificial intelligence needed reduce costs. The results implementing on Amazon dataset its comparison state-of-the-art models show superiority method.