The paper presents a semantic model of scientific thesaurus represented in the form of the domain semantic graph. Thematic scientific encyclopedias we use as a source of training data for model construction. Semantic graph clustering algorithms built on the basis of a thesaurus are proposed as a tool for aspect-oriented analysis of a scientific publication and analysis of a scientific field as a whole. The domain semantic graph «Mathematics» was constructed as a result of processing the text of the mathematical encyclopedia in five volumes. Examples of scientific publication analyzing experiments are presented. The task of dynamic aspect analysis is discussed. The proposed algorithm for aspect analysis of a scientific text is based on the transformation of the domain semantic graph into the publication semantic graph. The problem of automatic grouping scientific terms into groups corresponding to the different science branches is posed. The experiment results of domain graph clustering are presented.