<![CDATA[Music streaming platforms rely on playlists as medium for users to store musical preferences and receive music recommendations based on the music stored. However, representing playlists as meaningful groups remains a major challenge due to the high diversity characteristics of music. In addition, the distribution of musical characteristics within playlists can vary significantly. This study aims to compare two clustering models with different approaches hard clustering using the K-Means method and soft clustering using the Gaussian Mixture Model (GMM). Playlists are represented as statistical aggregations of audio feature data from songs, such as energy, acousticness, and danceability. The hard clustering approach using K-Means produces compact and clearly separated clusters, while the Gaussian Mixture Model (GMM) generates clusters that capture playlist ambiguity, resulting in overlapping clusters due to its probabilistic nature. These differences have a direct impact on the implementation of the clustering results in downstream applications. This study emphasizes the importance of selecting an appropriate clustering method for further implementations, such as music recommendation systems, and provides insights into the trade-offs between interpretability and flexibility offered by both models.]]>
Copyrights © 2026
