Advancing Unsupervised Learning on Graphs: Exploring Latent Structures of Graphs via Block Modeling