ABSTRACTS OF ARTICLES OF THE JOURNAL "INFORMATION TECHNOLOGIES".
No. 9. Vol. 26. 2020

DOI: 10.17587/it.26.523-528

A. S. Kruglik, Postgraduate, e-mail: kruglik.a.s@mail.ru, I. A. Lackman, PhD in Technical Sciences, Associate Professor, e-mail: lackmania@mail.ru, Ufa State Aviation Technical University

Hybrid Approach Content-Boosted Collaborative Filtering in the Field of Recommendation Systems

In this article, we conducted an empirical study of the predictive properties of hybrid recommendation system based on Content-Boosted Collaborative Filtering approach. The aim of the study is to improve the quality of forecasting the preferences of users of the developed recommendation system using of hybrid approaches. They overcome the disadvantages of individual approaches which predictions user preference. It is compared by four different quality metrics with other approaches of recommendation systems: random prediction, content-based filtering, collaborative filtering, averaging prediction. These approaches were tested on data on films and ratings provided by users. Content-Boosted Collaborative Filtering approach improves the result by 15—20 %, compared to other approaches. The usual method of collaborative filtering calculates the similarity of two users if and only if they rated one or more identical objects. Otherwise, if they rated different objects, then these users will be completely different. However, this proposal does not take into account that if the objects themselves rated by these users are similar, then these two users can be potentially similar. This method of finding potential similar users increases the number of nearest neighbors. In addition, the ranking of like-minded people can change significantly, therefore, the value of the predictions will also change in comparison with the usual CF method.
Keywords: recommendation system, hybrid approach, content-based filtering, collaborative filtering

P. 523–528

To the contents