We propose a graph decomposition algorithm for analyzing the structure of complex graph networks. After multi-word term extraction, we apply techniques from text mining and visual analytics in a novel way by integrating symbolic and numeric information to build clusters of domain topics. Terms are clustered based on surface linguistic variations and clusters are inserted in an association network based on their intersection with documents. The graph is then decomposed based on atom graph structure into central (non-decomposable) atom and peripheral atoms. The whole process is applied to publications from the Sloan Digital Sky Survey (SDSS) project in the Astronomy field. The mapping obtained was evaluated by a domain expert and appeared to have captured interesting conceptual relations between different domain topics.