Algorithm for Extracting Key Concepts from Educational Programs of IT Specialists using the Hybrid Context Ranking Method

In the context of rapid digital technology development and the growing volume of educational materials, ensuring interdisciplinary consistency in academic courses has become a critically important task for higher education institutions. The annual increase in digital data within educational systems underscores the need to develop effective methods for processing and analyzing curricula. Interdisciplinary consistency in an educational program involves creating coherent and logically sequential content across different courses, including identifying common themes, eliminating redundancy, and ensuring the correct sequence of material (e.g., introducing basic concepts before more complex ones). This consistency enhances the quality of IT specialist training, en­suring their competitiveness in the labor market, where professionals with comprehensive knowledge and skills are in high demand. A key step toward achieving this goal is automating the process of extracting key concepts from course syllabi. Key concepts are high-level ideas reflecting the core content of an academic discipline, which can be used to analyze connections between courses within an educational program. However, existing key concept extraction methods have significant limitations. For example, statistical approaches like TF-IDF, based on word frequency, cannot differentiate between different meanings of a polysemous term or account for semantic relationships between concepts. Graph-based methods like PageRank focus on structural relationships between words but often ignore their contextual meaning. These shortcomings are particularly evident when analyzing educational texts, where the same concept may be interpreted differently depending on the discipline or section of the syllabus. To overcome these limitations, a hybrid method called ContextualRank is proposed, combining semantic and contextual similarity analysis using pre-trained language models such as BERT and T5. This approach considers not only frequency characteristics and structural relationships but also the contextual usage of key concepts, making it more effective for analyzing course syllabi. The method uses a graph model with edge weights calculated through a combination of cosine similarity of vector representations and contextual metrics, along with the TextRank algorithm for concept ranking. To assess the relevance of key concepts to course objectives, a mechanism for analyzing direct and transitive connections was implemented using the Floyd—Warshall algorithm. Experiments conducted on data from Perm National Research Polytechnic University demonstrated that ContextualRank outperforms TF-IDF by 21 % in F-measure, achieving a precision of 0.7 and a recall of 0.93. Results were visualized as graphs highlighting key concepts and their connections to course objectives. The study demonstrates the potential of the method for automating the analysis of curricula, improving their structure, and adapting to dynamic labor market requirements.