On Graph Perturbation Theory And Algorithms For Scalable Mining Of Noisy And Uncertain Graph Data With Knowledge Priors.