DEVELOPMENT OF CONTENT RECOMMENDATION SYSTEM BASED ON AUDIOVISUAL CHARACTERISTICS

Abstract

This paper investigates the problem of creating effective recommendation systems for multimedia content using audiovisual characteristics. The research relevance is driven by the rapid growth of media content on the Internet and the need to improve the accuracy of personalized recommendations. Modern deep learning methods for feature extraction from media content are analyzed, including convolutional neural networks for visual characteristics analysis and recurrent neural networks for audio data processing. Approaches to combining heterogeneous content characteristics to form a unified vector representation are considered. The architecture of a hybrid recommendation system that combines audiovisual characteristics analysis with traditional collaborative filtering methods is proposed. A recommendation ranking algorithm based on multi-criteria optimization has been developed, taking into account both content similarity by audiovisual characteristics and user behavioral patterns. An experimental study of the proposed approach effectiveness was conducted on a large multimedia dataset. The experimental results showed a 15-20% increase in recommendation accuracy compared to baseline methods that only consider textual metadata. The practical value of the work lies in the possibility of applying the developed system to improve the quality of recommendations in streaming services, online cinemas, and music platforms.