One embodiment provides a method including: receiving input indicating at least one concept from at least one document of an input document collection; generating a validated data model for each of a plurality of concepts, wherein each of the concepts is represented as a visual data structure comprising semantics; generating at least one intermediate model object by parsing the validated data model, wherein each of the intermediate model objects comprises a concept object and the at least one intermediate model object identifies concept dependencies; translating the at least one intermediate model object into executable source code by importing at least one pre-built extractor and translating at least one rule identified from the visual data structure; and generating an executable information extraction program from the executable source code, wherein the generating comprises generating at least one rule for the executable information extraction program based on the identified concept dependencies.