This paper describes a customized scaffolding approach to improve learner's conceptual understandings through knowledge visualization. We propose a framework to use natural language processing and graph based algorithms to automatic visualize individual learner's prior knowledge states, domain knowledge, new encountered concepts and to reveal the semantic relationships between them. Thus, we are able to help learners to solve their uncertainties of new merging ideas and concepts in their learning process in order to integrate new knowledge with their preconceptions.