Machine learning model for musical emotion classification in recommender systems

The problem of classifying musical emotions arises when building recommendation systems for creating mood-based playlists. However, existing methods for classifying musical emotions do not always provide sufficient accuracy. This creates a need for the development of more effective methods and models, as many modern online music services strive to personalize content using recommendation systems, including the creation of mood-based playlists. To form such playlists, it is necessary to first solve the problem of recognizing the emotional tone of musical compositions based on their lyrics and audio features, which is a task related to the field of information management and processing. The aim of the article is to develop a machine learning model that improves the quality of musical emotion prediction based on the analysis of known and current methods and models. During the research, methods for extracting characteristics from musical compositions and techniques for organizing architecture to solve the problem were analyzed. An ensemble architecture and machine learning model based on stacking algorithms, multilayer perceptrons (with a ReLU activation function), decision trees, and bagging were developed for classifying musical emotions. A comparative analysis using the F-measure metric was conducted with alternative approaches on the same dataset. The developed model can be used in music recommendation systems for automatically creating mood-based playlists, improving user experience and the quality of personalized music services.