Active discriminative network representation learning

摘要: Most of current network representation models are learned in unsupervised fashions, which usually lack the capability discrimination when applied to analysis tasks, such as node classification. It is worth noting that label information valuable for learning discriminative representations. However, labels all training nodes always difficult or expensive obtain and manually labeling inapplicable. Different sets labeled model lead different results. In this paper, we propose a novel method, termed ANRMAB, learn active representations with multi-armed bandit mechanism setting. Specifically, based on networking data representations, design three query strategies. By deriving an effective reward scheme closely related estimated performance measure interest, ANRMAB uses adaptive decision making select most informative labeling. The updated then used further learning. Experiments conducted public verify effectiveness ANRMAB.