The purpose of this article is a theoretical consideration of the issues of digital simulation of the expert behavior in a given subject area, which essentially uses the processing of texts related to this subject area. As the main research task, the authors consider the issues of converting texts to a form that facilitates formal calculations and manipulations with them when creating an automatic expert system (ES) that functions as a digital simulator of the expert behavior. The article defines for the first time the content of the concepts of text Knowledge Bases for ES and the structural-typological model of the text. The article clarifies the features of using graph and network formats for representing text data extracted from primary texts for solving applied problems. It is proposed to use thematic text collections for the automated creation of ES Knowledge Bases. For ES, it is proposed to use the mathematical apparatus of Petri nets for analytical work with a graphical representation of texts.