This paper introduces a strategy and its theory proof to transform non-linear concept graph: Directed Acyclic Concept Graph (DACG) into a linear concept tree. The transformation is divided into three steps: normalizing DACG into a linear concept tree, establishing a function on host attribute, and reorganizing the sequence of concept generalization. This study develops alternative approach to discovery knowledge under non-linear concept graph. It overcomes the problems with information loss in rule-based attribute oriented induction and low efficiency in path-id method. Because DACG is a more general concept schema, it is able to extract rich knowledge implied in different directions of non-linear concept scheme.