Music recommendation model by analysis of listener's musical preference factor of K-pop

Abstract

Recently, the popularity of Korean pop music (or K-pop) has been increasing, due to technological developments in digital devices. According to the IFPI, 59% of Koreans regularly listen to music on their smartphones and other related devices [1]. Thus, the present study proposes a music recommendation model that predicts listeners' music preferences and recommends customized music lists. For this purpose, two pilot tests (with one participant in each test) were conducted and linear regression analysis was performed by using the Tensorflow application. The first test determined the participant's song preferences, based on a sample of 200 K-pop songs, while the second test added a neutral response option when considering a sample of 200 additional K-pop songs. The results indicate that the prediction accuracy of the participant's song preferences in the first test was 71.5%. However, after adding the neutral response option in the second pilot test, the prediction accuracy increased to 84.0%. This model can be used to predict the music preferences of listeners on a wider scale.