Факультет комп'ютерних наук (КН)

Постійний URI для цього фонду

https://openarchive.nure.ua/handle/document/43

Перегляд

Перегляд Факультет комп'ютерних наук (КН) за автором "Alokhina, M."

Зараз показано 1 - 1 з 1
  • Зображення мініатюри
    Публікація
    Determining preferences in recommender systems based on comparator identification technology
    (ХНУРЕ, 2022) Beskorovainyi, V.; Kolesnyk, L.; Alokhina, M.; Кosenko, V.
    The subject of research in the article is the process of ranking objects in the lists of recommender systems. The goal of the work is to increase the efficiency of recommender systems by improving the method of determining preferences between objects in lists using the theory of multi-criteria decision-making. The following tasks are solved in the article: review and analysis of the current state of the problem of identifying advantages between objects and their ranking in the lists of recommender systems; analysis of filtering methods used in recommendation systems; decomposition of the decision support problem for selection of objects; development of a combined method for ranking objects in the lists of recommender systems, combining the procedures for selecting a subset of Pareto-optimal objects, structural-parametric synthesis of a scalar multi-criteria estimation model, and evaluating the entire set of selected objects. The following methods are used: mathematical modeling, systems theory, utility theory, decision theory, optimization and operations research. Results. Based on the results of the analysis of the modern methodology for ranking objects in the lists of recommendation systems, the possibility of increasing their efficiency has been established. To take into account factors difficult to formalize, the knowledge and experience of users, it is proposed to implement the determination of preferences between objects using the theory of multi-criteria decision making. The problem of forming lists of recommendation systems is decomposed into the tasks of selecting a subset of Pareto-optimal objects, structural-parametric synthesis of a scalar multi-criteria estimation model, and evaluating a set of selected objects. A combined method for ranking options has been developed that combines the procedures of ordinalistic and cardinalistic ordering technologies and allows one to correctly reduce the subsets of objects included in the lists of recommendations. Conclusions. The developed method for determining preferences expands the methodological foundations for automating the development and operation of recommendation systems, other multi-criteria decision support systems, allows for the correct reduction of the set of non-dominated objects for the final choice, taking into account factors that are difficult to formalize, knowledge and user experience. The practical use of the obtained results due to more economical method of forming lists when adding new objects will allow to decrease the time and capacity complexity of the procedures for providing recommendations, and due to taking into account of set of weighted local indexes and allocation of set of non-dominated objects - to increase quality of given recommendations.