DATA REPRESENTATION MODEL FOR A RECOMMENDATION SYSTEM IN THE EDUCATION FIELD BASED ON FUZZY LOGIC

Abstract

Analysis of modern research in the field of development of recommendation systems showed that they can be used quite successfully in the educational field. At the same time, the quality of the recommendation largely depends not only on which approach to building the recommendation is used, but also on how the data are presented and which of them are taken into account in the recommendations. The paper provides a rationale for choosing a data representation model based on fuzzy logic. When building models of fuzzy variables, the context of the domain of the subject area is taken into account, namely: the types of possible recommendations are determined; term-sets corresponding to the semantics of parameters and recommendations are formed; sets of alternative term sets are determined using the example of determining the discipline rating. Data modeling was carried out using triangular and Gaussian membership functions depending on the power of term sets of fuzzy variables: triangular or truncated triangular functions were used for term sets corresponding to a non-binary scale, and Gaussian membership functions were used for binary features. The issue of multi-criteria rating indicators is considered and an example of evaluating a discipline based on several indicators, which are components of its rating, is given. Data modeling was carried out and a vague conclusion was formed regarding the recommendation of the discipline using the Mamdani method. An example of the extension of the proposed approach to building a data model to other indicators of the recommendation system is provided, and a list of such indicators is determined according to the subject field of education. The model allows to take into account indicators that are a number in a certain range (at the same time, both discrete and continuous scales can be used) or a logical (binary) value (the interpretation of each value is determined by the context and can be interpreted in different ways in each individual case).