Application of the descriptor approach for clustering entities from education sector

Abstract

The article proposes the algorithm to solve objects clustering problem for such subject areas as education and labour market. Such objects are competence, discipline, specialty, vacancy, etc. The main problem in clustering algorithm development proved to be the stage of attributes design since the named objects have descriptions in a natural language. Consequently, a descriptive model for the objects was designed at first. The model was based on the fact that all necessary concepts are characterised in the space of descriptors ”know”, ”can”, etc. This allowed the object to be represented as a tuple based on the object name, descriptors (and their values) and keywords related to descriptors. To obtain such structures, the toolkit of context-relative text mining was used. The ability to represent the entities in question as a formal structure allowed the attribute space formation algorithm to be developed and complex metrics to be constructed to solve the stated clustering problem. The developed algorithm permits development of various services for a faster and more objective decision making process in educational and professional sectors. As a result of the work, about two thousand vacancies were obtained and transformed into descriptor entities. Based on the error matrix, it can be judged that the resulting descriptor entities have been clustered with a sufficient level of quality. This demonstrates the applicability of the model for presentation and analysis for elements from the subject area.

Assessment of the balance of the competence formation process: analysis of the curricula of Russian universities

“…We agree with the opinion of authors [14], that the quality of learning outcomes is influenced by various factors, primarily existing curricula, educational programs and standards, in particular the structure of the curriculum. To optimize the structure of the curriculum, many researches build graph or network models [4] applicable by well-known and tested graph theory algorithms [1], fuzzy logic methods [5], or neural network approaches [22]. However, the input data for these models are, mostly, the subjective opinions of experts evaluated, for example, on the Likert scale [20], which reduces their adequacy.…”