This paper proposed a modified algorithm of grey structural modeling (GSM). In the conventional algorithm, a hierarchy parameter was sensitive to the hierarchical structure of a digraph, which makes it difficult to find out the proper structure of a digraph.In the proposal, the number of hierarchy could be directly determined by the K-means method with factor analysis support. Theimpression of twenty table lamps with ten adjective-word pairs was measured for nine university students by the semantic differentialtechnique and the impression data were analyzed by factor analysis, statistical graphical modeling, the conventional model and theproposal. As a result, the conventional algorithm was difficult to adjust the hierarchy because of the narrow range of the parameter, but the proposal was able to determine it. The proposal would contribute to establish a more valuable way for modeling kanseistructure even when there was little experience