Towards the Identifiability and Explainability for Personalized Learner Modeling: An Inductive Paradigm