Recommender systems: a survey of modern approaches

Abstract

This paper is a survey of algorithms that are used in recommender systems. All related methods may be divided into three large groups: collaborative filtering, content-based and knowledge-based methods. These approaches are based on different source data that is used for recommendations computing. Collaborative filtering is based on analyzing ratings that user provide for items. The idea here is to recommend items that got high rating by the users that are similar with the user, who receives recommendations from the system. Content-based methods use available content of recommended items. Predominantly the content is represented as texts from items' names, descriptions and so on. Methods that analyze media content (images, videos, audio) are not considered in this paper. Knowledge-based methods analyze user requests that include information about what items do they want. The most popular methods are considered in this paper. Advantages and drawbacks of recommender technics of each group of methods are described. Moreover, composite techniques that allow drawbacks to be eliminated are considered in the paper. Composite technics assume that a lot of information about items is available. Hence, they combine collaborative filtering, content- and knowledge-based technics to provide more accurate recommendations for users.