Avaliação da influência de características musicais sobre a popularidade de canções nas charts do spotify brasileiro em 2021

Abstract

Currently, the digital format is the main music’s distribution channel and the spotify platform stands out as the main distributor. The revenues generated by the songs are shared between distributors, artists, composers and record studios, being proporcional to the number of streams. In this scenario, distributors that adopt methods capable of predicting song’s popularity before their release are more competitive as they can optimize the allocation of financial resources to songs that are more likely to reach high popularity. This work have the main goal to the influence of musical characteristics on the performance of songs on Spotify’s Brazilian charts in 2021, using random forest classification model and interpretability techniques of black box models such as features importance, permutation importance and SHAP feature Importance. Other techniques not dependent on machine learning models such as pearson’s correlation and information value are also applied. Data regarding the number of streams, artist characteristics and musical characteristics of songs that reached the top 200 of brazilian Spotify charts in 2021 were collected and a popularity index was calculated for labeling the songs between popular and common songs. The results show that artist characteristics such as the number of followers, the artist’s popularity index and the musical genres assigned to the artist are the three most important characteristics for determining the popularity of a song, followed by musical characteristics such as speechiness, acousticness, energy, tempo and liveness.