A research team at Cornell has come up with an interesting finding about music streaming platform Spotify. According to them, the platform, which relies on their users’ likes to predict what song they’d prefer to listen to next, and recommending the same to them, would do well to incorporate a Dislike button into its interface. The results of their study say that adding a Dislike button will help Spotify recommend better.
Dislikes Matter Too
The research team has come up with a recommendation algorithm that allows them to predict how incorporating both likes and dislikes into its platform will help make Spotify more effective. Their study has shown that listeners, on and average, are 20 percent more likely to give a song a “like,” if the recommendation algorithm is developed using 4000,000 likes and dislikes, as opposed to an algorithm that is trained using only the number of likes.
Senior Research Associate Sasha Stoikov, from the Cornell Finanical Engineering Manhattan, who also happens to be the lead author on a paper about the algorithm, believes that if the algorithm is based only on likes, then it may help users discover new and enjoyable songs, but at the same time, it is also more likely to recommend songs that the users won’t like.
Piki to Help People Pick
The paper, which has been titled, “Evaluating Music Recommendations With Binary Feedback for Multiple Stakeholders,” was first published by Stoikov and Hongyi Wen on September 15, and is all set to be presented at the ACM Conference on Recommender Systems.
The algorithm that was used in the study has been nicknamed “Piki” (read, “picky”) by the researchers, and apparently makes use of a database spanning some five million songs, luring users into rating the songs by offering them $1 for every 25 songs that are rated by them. The interface first plays a song, giving users the opportunity to rate it after different time intervals. “Dislike” is allowed after the first three seconds, “like” after six seconds, and “superlike” after 12 seconds. Stoikov says that the incentives make it possible for them to have users “vote truthfully.”
Problematic to the Artist
Apart from not being the best music recommender to listeners, the researchers have found another flaw with Spotify’s algorithm. They say that it makes it harder for lesser-known artists to be discovered, since their songs are not recommended easily due to lesser number of likes. This discrepancy, they hold, favors well-known artists.