Many different versions of the same song are often found in online music or video services. To enable easier search and better user experience, online music services often use clustering techniques to cluster different versions of the same song. Clustering systems are designed to be tolerant to shifts, noise, small tempo changes, etc. This tends to cause remixes and covers of a song to be clustered with the original versions of that song. This leads to certain problems, e.g., recommender systems selecting the most popular version of a song, in preference to other versions of the song; poor precision-recall rates of song metadata; etc. In turn, these lead to poorer experience for both users and creative partners of the music service.

This disclosure describes techniques that group music or videos into sub-clusters of increasing dissimilarity, based on signals that indicate provenance, quality, popularity, officiality, etc. The techniques enhance user experience by returning more appropriate search results.

Creative Commons License

Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.