IMPLEMENTATION OF A HYBRID METHOD OF SEARCHING FOR CLOSE OBJECTS, TAKING INTO ACCOUNT THE GENERAL AND ACOUSTIC CHARACTERISTICS

Abstract

The subject of research in the article is the methods of finding close objects and technologies of forming recommendations. The aim of the article is to develop a recommendation system based on a hybrid method of searching for objects, taking into account both user preferences and audio characteristics of objects. The following tasks are solved: analysis of methods and algorithms used in recommendation systems; development of a hybrid method of forming recommendations on the principle of double organization; determination of the main functions and architecture of the system of formation of musical recommendations; testing of calculation algorithms and search methods in the system for analysis of similarity of musical recommendations. The following research methods are used: methods of correlation analysis, methods of similarity theory, algorithms of collaborative filtering and content analysis, hybrid methods, methods of analysis of audio characteristics, programming technologies. The following results were obtained: A study of collaborative filtering, content-based filtering and hybrid methods. Algorithms and calculation formulas of the considered methods are given. The main audio characteristics of musical compositions are considered. The method of formation of recommendations on the principle of double organization is developed. The main functions of the system of formation of musical recommendations are listed and the diagram of components is formed. An example of calculating the characteristics of user preferences and similarity of musical compositions by audio characteristics is given. Conclusions: According to the results of testing the system by three methods, we can conclude that the proposed hybrid method was the most effective among the studied recommendation methods with the lowest standard error rate. In addition, the hybrid method on the principle of double organization solves such problems of existing recommendation methods as excessive similarity of recommendations, potentially small number or no proposals at all by compensating data from one block of data from another.