Backward-chaining rule induction is a process abstraction, a semi-supervised, goal-informed strategy of constraining the search for (what amount to) association rules, ideally ones that are most relevant to an overarching goal or task, by embedding a supervised rule induction engine into an iterative exploratory loop. This paper also discusses interfacing induction with prior knowledge -- a curated knowledge base of molecular interactions that can be used to assess hypothesis "interestingness"

Fisher, D., Edgerton, M., Chen, Z., Tang, L., & Frey, L. (2006). "Backward Chaining Rule Induction," Journal of Intelligent Data Analysis, 10, 5, 397--417.

Bootstrapping rule induction is a process abstraction, a method of adapting an off-the-shelf rule induction engine to form stable rules with condition and outcome uncertainties, all of which makes the rules more believable to experts.

Waitman, L. R., Fisher, D., & King, P. H. (2006). "Bootstrapping rule induction to achieve rule stability and reduction," Journal of Intelligent Information Systems, 27, 49--77.