Method of Generating Recommendations Lists With Considering Activity Indexes of Users in a Recommendation System

Abstract

The subject matter of the article is the process of generating recommendations lists for users of a recommendation system. The goal is to develop a new method of building recommender systems to improve the quality of recommendations lists, increase user space coverage and item space coverage with considering information about user activity indexes; development of a hybrid of this method with the method of collaborative filtering. The tasks to be solved are: to develop the method of building recommendation systems based on considering activity indexes of users, develop software to test this method, conduct experiments with the developed software to test the effectiveness of the developed method, determine the quality of its work and compare this method with the standard method of collaborative filtering. The methods used are: graph theory, mathematical statistics, the theory of algorithms, object-oriented programming. The following results were obtained: the expert-oriented method for building recommender systems based on considering indexes of user activity and calculating expert coefficients has been developed, the hybrid of this method with the collaborative filtering method has been developed, software for implementing this method and this hybrid has been developed, experiments with developed software to test the developed method and the hybrid has been conducted. Conclusions. The possibility of using information about user activity in recommender systems to improve the quality of recommendations lists has been investigated. The calculation of expert coefficients is proposed to supplement the similarity coefficients in recommender systems. An expert-oriented method for constructing recommender systems based on considering activity indexes of users and its hybrid with collaborative filtering has been developed. Experiments has been conducted with the developed software have shown that the developed method significantly improves such indicators of the recommender system as user space coverage and item space coverage and allows to create higher quality of recommendation lists without significant fluctuations Precision and Recall of the recommender system, and in some cases even improve these indicators, it depends on the features of the input data. K e ywor d s : recommendation systems; collaborative filtering; similarity coefficients; expert-oriented filtering; expert coefficients.