Articles
p. 4-13
Mathematical Modeling and Forecasting of Music Popularity: an Overview of Approaches and Methods
The article presents a comprehensive review of modern methods of study of musical compositions popularity, including statistical, temporal and connectionist approaches. Regression models with fixed effects, survival models, nonstationary Poisson processes and systems of Lotka-Volterra equations, as well as STL decomposition algorithms and neural networks (LSTM-RPA) are considered. Their advantages and limitations in analyzing the dynamics of listening, the influence of playlists and competitive interaction of genres are discussed. The main attention is paid to the analysis of their advantages and disadvantages in the context of the tasks set - forecasting long-term changes in the popularity of compositions, assessing the impact of playlists and the dynamics of genre competition. On the basis of the analysis, the author concludes that the use of LSTM-RPA models is the most effective for forecasting due to their ability to minimize errors accumulation. The author contribution is in proposition of a new approach: the adaptation of hybrid methodology combining neural network models and time series for taking into account contextual factors such as influence of social trends and marketing activity that can significantly increase the versatility and accuracy of the forecasts.
