On the Scalable Learning of Stochastic Blockmodel