Interpretable Network Representation Learning with Principal Component Analysis