Does the 'Not good!' button have any meaning?
In today's world, where there is so much content that it would be impossible to consume it in a lifetime, the recommendation function that recommends the most suitable content for each user is playing an active role everywhere. Especially in music streaming services like Spotify, it is an important function that users will leave depending on the superiority of recommendations. There is no low-evaluation button in such Spotify, but researchers at Cornell University have summarized the difference in accuracy between 'high-evaluation button only' and 'high-evaluation button & low-evaluation button' in a paper. Did.
Evaluating Music Recommendations with Binary Feedback for Multiple Stakeholders by Sasha Stoikov, Hongyi Wen :: SSRN
'Dislike' button would improve Spotify's recommendations | Cornell Chronicle
https://news.cornell.edu/stories/2021/09/dislike-button-would-improve-spotify-recommendations
The experiment was conducted in the form of asking users to listen to music selected from 5 million songs and evaluate it. According to Sasha Stoikoff, who led the experiment, the first 3 seconds of the song cannot be rated, 3 seconds after starting to listen to the song, ``low rating'' can be selected, 6 seconds later, ``high rating'', 12 By making it a format in which the option of 'super high rating' can also be selected after seconds, he said that he was careful so that high evaluation can be selected only when the user really likes it.
Based on the 400,000 data collected in this way, the algorithm was trained with only 'highly rated & unrated' data, excluding low rated data, and 'highly rated & low rated' data was excluded. We examined the difference in accuracy when training only on the data of . Then, the songs recommended by the algorithm trained on 'highly rated & low rated' data are 20% more likely to be highly rated than the songs recommended by the algorithm trained on 'highly rated & not rated' data. It turns out.
In the experiment, Mr. Stoikoff also investigated the relationship with name recognition, and found that lesser-known artists are less likely to receive high ratings and are less likely to be recommended by many people on platforms such as Spotify. In addition to the fact that lesser-known artists are less likely to receive high ratings, many algorithms are trained based on which songs they have listened to, rather than user ratings, so they are already well-known and listened to by many people. Stojkov says that only artists who are recognized tend to be recommended.
According to Stoikoff, the ultimate goal of the research team is to work with record companies to ``be able to determine how likely users are to like a song before distributing it to a big platform.'' . Future research will focus on other distribution platforms such as Netflix, incorporating more advanced algorithms, and modeling how the goals of stakeholders such as platforms and record labels are linked to those of the algorithms. said the person.
Related Posts:
in Note, Posted by log1d_ts