I’m baffled by the fact that you can’t thumbs up or down the song you’re listening to but only the song coming up next. I listen to radio in part to discover new artists/tracks. So more often than not I don’t even know the next track and have no basis to thumbs up or down. This feature doesn’t make much sense to me.
If you skip the current song, Roon will ask why.
1 Like
Yes, I get that, but not sure it has the same effect on future behavior. I guess the real question what is the result of giving the next song thumbs down vs skipping forward from the now playing song and giving feedback that ‘I don’t like this song’.
Has been answered by Roon before, searching the forum will find older discussions. Essentially it’s a tiny weighting being added to the many other inputs for Valence. It won’t mean that you never get a song by that artist in Radio anymore or something like that. Don’t overthink it 
If you click any link in a Google search result, it also adds a tiny weighting to the overall statistics.
1 Like
The thumbs up/down is simply a bad metaphor because it’s commonly associated with like/dislike.
Well, even if some others symbols would be used, in my opinion it is really completely useless. Who wants to listren to “radio” and then watch the mobile and decide whether next song should be played or not? I cant imageing anybody is doing this
Therefore, logical is and what I would expect of a high priced service as roon is, what Remco_K said
During Roon radio, on the bottom of the screen there’s a box showing what track is coming up. There I can like or dislike it. But that’s mostly not useful because for me Roon Radio is for discovery of new music that I haven’t heard before.
So it would only be logical if I could like or dislike a song currently playing.
I dont understand why Roon learns from Spotify, the industry leading service for learning what users wants to listen and a great radio function.
For me, as a roon user since a few months this is quite disappointing.
1 Like
You don’t have to watch it continually. You can click the button many times and build the queue in advance, leaving out any tracks you do know and don’t want on the queue. I, for one, do this occasionally.
How is Roon going to know if you ever heard a song before in a friend’s car or whatever? If it’s playing songs that it thinks you like, occasionally there will be one you know already.
Personally, I don’t want Roon Radio to play only songs I never heard, anyway.
You can by skipping it, then it asks why you skipped it as explained above
I have the same trouble with Roon Radio(Android Remote APP). Users need to make a “like” or “dislike” evaluation before hearing music they have never heard before, which does not conform to the product logic. Moreover, when the next song is about to be played, the user may not be on the roon interface, but may be reading a novel, browsing the community, etc. When they hear a track they don’t want to listen to, it is too late to press “dislike”.Roon Radio thus will always recommend the disliked B after A.
3 Likes
Well, not trying to overthink this but simply use some logic and try to understand. It seems to me that if Valence is trying to learn our preferences it would be helpful if we could give the thumbs up input - i.e. we especially like this track. When we go to the next track we are only giving negative feedback about a particular track.
I don’t know how the the learning works exactly, I don’t think it was fully disclosed by Roon as far as I can tell, but there was one post by Roon staff with some details that I can’t find now. To me it seems that there might be 2 positive inputs: You add the track to the queue explicitly with the thumbs up button and/or you don’t skip it when it plays. I am not a machine learning expert at all but maybe it’s more important in a radio situation to avoid playing bad tracks.
Anyway, this has been discussed for years and I am sure you will find everything that can be said along with some Roon staff explanations in one of the many older threads:
https://community.roonlabs.com/search?q=roon%20radio%20thumbs%20
However the actual algorithm is working, it would be helpful for the training of the algorithm if I can like / dislike the currently played song as it is a stronger message from compared to add or not add the next (to me completely unknown track) to the queue.
With all the current ChatGPT fuzz out there, I am pretty sure expectations of peoples for such things will rise.
1 Like
IIRC, individual likes and dislikes do not have a significant impact on the algorithm; machine learning uses many millions of data points, so individually we’re not going to influence things very much at all.
Also, I suspect the positive action of not skipping a radio track, i.e., listening without interacting, is certainty more frequent than skipping a track and selecting a reason for doing so.
1 Like
You could do a weighing. So deliberately liking a song or not is more relevant than listening through a song without skipping.
To be clear, I’m a user like you. However, my point is that individually liking a song–even if weighting was used as you suggest–would not change things directly for you. Because one data point from many, many hundreds of thousands of listening hours is literally a drop in the ocean. Valence learns from collective listening habits, and uses this to offer tailored listening suggestions.
But if you don’t know the next song, which is inevitable on radio and probably one of the points of radio, you can’t say anything about it anyway. If it then plays and you hate it, you would probably skip it. If you don’t, you probably don’t hate it enough. And like @Martin_Webster says, any action is just at tiny weight in millions
1 Like
Ah, sorry, took me a while. I thought Roon Radio learns primarily from my listening habits.