Modeling self-efficacy in intelligent tutoring systems: An inductive approach