The development of a hybrid recommendation system

Recommender systems are programs for finding objects that respond to user’s interests. Most often, recommender systems are used in Internet networks for commercial purposes. To attract users to the website, the recommender system is adapted to their tastes and preferences. "Intelligence-of recommendations allows users to find information that interests the them. Based on the type of objects, the audience coverage, the method of obtaining information about objects and users, as well as the required accuracy and completeness of the recommendations, methods and algorithms on which the system of recommendations is based are selected. The most common methods include statistical methods, collaborative filtering techniques, content filtering methods, contextual methods and hybrid methods. Each of these methods has its advantages and disadvantages. Some of the problems that are involved with developing recommender systems are difficult to solve within a single method, so hybrid recommender systems are of interest. The optimal combination of methods allows to develop the most effective system of recommendations. For example, combining collaborative filtering and content filtering takes advantage of these methods individually. Content filtering compares the content of objects with the interests of the user, but does not take into account the quality of these objects. In collaborative filtering recommendations are made depending on the ratings of users ratings (estimates). Therefore, the user will not be offered objects with a low rating. Also, collaborative filtering gives the user atypical objects that are not included in the list of interests of the user, but could please him. Thus, the problem of limiting the range of interests of the user is solved. The content filtering method, in turn, solves the problem of new objects. If the object has not been rated by any user yet, it will not be included in any list of collaborative filtering recommendations. However, if you add an object to the database and specify some information about it in the form of text (or if the object itself is text), content filtering will be able to identify the most similar objects and recommend it to users who chose such objects. The purpose of the work: to identify the main problems of recommender systems and methods of their solution; to develop a hybrid system of recommendations, using as the basic methods of collaborative filtering and content filtering; using metrics of accuracy and completeness, to determine the most effective method of composition methods.