Khudik B.O. Methodology for building a recommendation system for the field of education based on a hybrid approach. – Qualifying scientific work on the rights of the manuscript.
Dissertation for the Doctor of Philosophy degree in the specialty 123 Computer Engineering. - State University of Information and Communication Technologies of the Ministry of Education and Science of Ukraine. - Kyiv, 2023.
Modern approaches to the organization of educational processes in the formation of the individual educational trajectory of education seekers require the formation of various recommendations, which makes these processes the sphere of interest of recommendation systems. Analysis of the use of recommender systems showed their significant focus on ensuring commercial tasks and goals, while in the field of education, the primary task is the quality of the educational process. Also, there is a significant difference in the objects of interest and the target audience of recommendation systems in the educational field: a clearly defined audience of users; one-time use and fleeting relevance of most recommendations; the opportunity to obtain preliminary assessments of objects of interest due to mandatory surveys. This determines the need to create a separate methodology for building a recommendation system in order to improve the quality of solving various tasks of supporting decision-making in educational processes.
The work solves the actual scientific and applied task of improving the quality of providing recommendations in the field of education through the use of a recommendation system based on a hybrid approach.
An analysis of modern methods, models and approaches to building recommender systems was carried out, in particular: content filtering; collaborative filtering; hybrid approaches. An overview of expert evaluation methods was conducted, the specifics of defining and using different scales were considered, and the issue of forming generalized expert evaluations for calculating the general rating of objects of interest to users was considered.
Since a significant part of the characteristics has a linguistic nature, it is proposed to use a data presentation model based on fuzzy logic, which allows for a simplified transition between numerical rating scales without changing the linguistic presentation.
The application of the methods of recommendation systems in the field of education is considered in the context of the most typical tasks of applicants that provide support for the educational process.
The developed method of generating recommendations for objects of interest using a hybrid approach can be used to build recommendations taking into account the following features: the system knows the user's object of interest, the objects of interest contain a certain textual description; the degree of relevance of the object of interest to the user's request takes into account the semantic proximity of the request (or the user's characteristics) to the objects of interest, and in the next steps - user evaluations, information about the interaction of the acquirer with the object of interest and his behavior in the system. To assess the relevance of the user's interests to the recommendation objects, it is suggested to use natural language processing methods, in particular, for preliminary filtering of the objects of interest. This allows to reduce the size of the vector of potentially useful objects. The use of targeted surveys allows reducing the size of collaborative filtering matrices at the last stage of recommendation formation. Also, the rating of objects of interest based on targeted surveys allows to reduce the problem of "cold start", since the number of objects without rating information in the system is very low, and even for such objects it is easy to obtain a preliminary assessment based on similar assessments of similar objects with them. Approbation of the methodology was carried out by developing a recommendation system to solve the problems of forming a recommendation regarding the choice of the discipline of the student's free choice block and the selection of educational literature. The response of the model is 92% - so many users agreed with the recommendation provided by the system. The level of satisfaction of users with the received recommendation for choosing a discipline over time is 76%, while 15% would like to change their choice, but to those disciplines that the recommender system provided them at the top of the list of recommendations. The study of the level of relevance of the recommendation when solving the task of selecting literary sources showed an average rating of 8.5 on a 10-point scale.
Keywords: recommendation system;educational sphere;automation;information technology;information system;decision support;mathematical modeling;database;fuzzy logic;fuzzy model;user activity results;machine learning;natural language text processing;hybrid approach.
