Recurrent Attention Walk for Semi-supervised Classification

Abstract

In this paper, we study graph-based semi-supervised learning for classifying nodes in attributed networks, where nodes and edges possess content information. Instead of treating all neighbors uniformly, we learn how to explore neighborhoods through a reinforcement learning walk policy tuned for classifying unlabeled target nodes.

A key graph learning contribution from the KAUST research period, published at WSDM 2020.