You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
It would be great if there is a support for using negative signals as inputs to the model. In the context of a music-streaming platform, a song skipped by a user is a negative signal which can be useful for understanding the preferences of a user.
Motivation
Negative signals, just like the positive signals, contain useful information about the preferences of a user, and being able to use it for training recommender systems can be helpful in delivering enhanced experience to the users. An example of a positive signal could be downloading a song. And an example of a negative signal could be skipping a song within 30 seconds. Using skip signals have shown to improve the user engagement as mentioned in this paper by Spotify.
The text was updated successfully, but these errors were encountered:
@animesh-wynk can you give more details how these negative signals are defined given a positive? Meaning if someone downloads a song now, but skip another song two hours later, are you using this skipped song as negative signal for the song downloaded two hours ago? Basically how positive-negative pairs are integrated? thanks.
Hi Ronay!
If a song is streamed for >= 30 seconds, then we count it as a successful
stream, else it is considered an unsuccessful stream as it was skipped too
soon. So a session can have an array of consumed songs and an array of
skipped songs. One of these 2 arrays can be negative as well, in case all
the songs in that session were consumed or skipped.
Feel free to ask me more about it if you still have any other questions or
doubts.
Thanks
Animesh
On Wed, Jun 28, 2023 at 7:08 PM rnyak ***@***.***> wrote:
@animesh-wynk <https://github.com/animesh-wynk> can you give more details
how these negative signals are defined given a positive? Meaning if someone
downloads a song now, but skip another song two hours later, are you using
this skipped song as negative signal for the song downloaded two hours ago?
Basically how positive-negative pairs are integrated? thanks.
—
Reply to this email directly, view it on GitHub
<#702 (comment)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/AQVXDVJ65DHEFUG52KRRN2DXNQXTXANCNFSM6AAAAAAYDJAE2I>
.
You are receiving this because you were mentioned.Message ID:
***@***.***>
🚀 Feature request
It would be great if there is a support for using negative signals as inputs to the model. In the context of a music-streaming platform, a song skipped by a user is a negative signal which can be useful for understanding the preferences of a user.
Motivation
Negative signals, just like the positive signals, contain useful information about the preferences of a user, and being able to use it for training recommender systems can be helpful in delivering enhanced experience to the users. An example of a positive signal could be downloading a song. And an example of a negative signal could be skipping a song within 30 seconds. Using skip signals have shown to improve the user engagement as mentioned in this paper by Spotify.
The text was updated successfully, but these errors were encountered: