The Conundrum of Success in Music: Playing it or Talking About it?

Abstract

Nowadays social media are the main means for conducting discussions and sharing opinions. The huge amount of information generated by social media users is helpful for predicting outcomes of real-world events in different fields, including business, politics and the entertainment industry. In this paper, we studied the possibility of forecasting the success of music albums by analyzing heterogeneous data sources spanning from social media (Twitter, Instagram and Facebook) to mainstream American newspapers (e.g., New York Times, Rolling Stones). The idea is to exploit music albums' pre-release hype and post-release approval to predict the album's rank with reference to the well-known Billboard 200 album chart, which tabulates the weekly popularity of music albums in the USA. To predict the success of a music album, that is its rank in the chart, we identified metrics based on the messages' posting trend, the variation of the sentiment associated to such messages, the number of followers of the album's author, and the importance of the people who talk about it. To evaluate the effectiveness of the proposed metrics we have compared the prediction performances of several models based on supervised learning approaches among those most used in literature. As a result, we obtained that the Random Forest approach is able to predict the music album rank in the Billboard 200 Chart with an expected accuracy of 97%. As a further validation, using this specific model, we also conducted an additional real usage test obtaining an almost matching result (accuracy of 94%).

Keywords

• Social media
• machine learning
• prediction
• sentiment analysis
• music industry