Roon explains that recommendations get better over time the more you use Roon.
I have been using Roon for just about a week and at the moment I am dissatisfied with how recommendations work. I’ve seen similar opinions in the community.
That’s why I wanted to ask people who have been on Roon for a longer time, are the recommendations genuinely getting better and more relevant over time, and are you satisfied with them?
If so, approximately how long did it take before the proposals in the recommendations improved?
Valence does more than compare you to other users. It uses Roon’s deep metadata as its foundation, providing context about music and connections between the people who compose, perform, and produce it. Next, it considers your listening history before filtering for the results that are most popular, but only among listeners who are experts in that genre.
I was just interested if there was a statement because I don’t think I ever saw info on how the recommended albums are selected.
I’m also wary about the info on Radio in your link because there are various statements by Roon staff on the forum that the feedback created through Radio has no immediate effect on the personal Radio recommendations and is a drop in the bucket of the big cloud.
That said, personally I am not a big Radio user but if I do I have no qualms with it and am not experiencing eternal repetitions or other issues that other people mention. It seems to do its job for my needs quite well. (Although I don’t think that an AI can ever be as good a DJ as a human who knows what they are doing)
I agree, I haven’t had repetition or duplication issues with recommendations. My listening is usually my own prescription or based on community recommendations, I rarely let Roon take over.
For me Roon’s recommendations vary quite a lot, from useful to useless, depending on the album and genre. I expect to some degree it also depends on how well I know a particular genre myself.
In terms of it getting better over time, that’s not something I can say I have noticed myself having used Roon for a few years now, but that’s not to say it doesn’t.
The Valance marketing materials would seem to suggest it does, but be good to hear an official statement from Roon on how much.
I don’t get the feeling that Roon knows that I am more of an expert in certain genres and hence goes ‘deeper’ on those genres, with more ‘basic primer’ suggestions for genres I listen to less. But again, be good to know if that is expected behaviour or not.
Roon Radio is a separate subject, and probably better left to another thread, but for me, basing future selections solely on the last track played severely limits its usefulness. If I am listening to a playlist made up of a mix of genres and the last track happens be a classical composition, that doesn’t mean I want to listen to another hours worth of classical music. Conversely, in general, Roon’s album recommendations ‘seem’ more relevant by comparison because it’s clear they only relate to a specific album or genre.
Indeed, same for me with my genres. Though I don’t think that Roon Radio is super bad at this - there is some variation for me, which I appreciate.
(I don’t understand the many complaints that Radio does not stay closely in the start genre: Like you, just because I start Radio from a Punk or Surf song does not mean that I want exclusively punk or surf for hours and hours.)
That’s why I don’t have much hope in AI DJs generally. I’d prefer that a punk song with a polka rhythm is followed by some old German folk song with a polka rhythm, rather than yet another punk genre song. Or that a surf song about girls is followed by another genre song about boys with a similar topic. What a good human DJ would do …
Though yeah, it could probably be improved it somehow it was possible to seed the Radio with a selection of genres/artists/tracks and not just one artist, genre, or the last track in playlist.
In most streaming apps, “New releases” are just a list of everything that’s been released this week, or sometimes only certain titles that are being marketed. Valence lets Roon show you new releases you care about, based on your library and listening history. The more you use Roon, the better the “New Releases” will get!
Valence is a system built on a massive new cloud database, which uses machine learning algorithms…
“Machine learning algorithm” suggest that it learns and ges better in time too.
Thank you, I had forgotten about that. As some wrote above, I can’t say either if it got better for me over the past 2.5 years of using Roon, but generally the recommendations work quite well for me and they make sense in the context of my library and listening
I find that the recommendations in the “New releases for you” section of the Home page are pretty good for me and tuned to my listening habits and history. Here’s the latest serving, and while it inevitably has Pink Floyd, I’m pleased to see that Valence knows about my fondness for Brian Eno, Martha Agerich, Enigma, the Pet Shop Boys and countertenors…
Better is a loaded statement. I would say, what you listen to influences recommendations.
What is in your play history definitely affects suggested content. Now, how much and how quickly play history affects things are up for discussion. Personally, I think that it takes quite a bit to nudge Valence, hence the “more you use” comments.
For example, and I’ll use my own experience. I left an album playing on repeat, turned off the system, but didn’t stop Roon’s playback. I then went on a two week vacation while that album was playing on repeat. Years later, as Roon released various recommendation systems, I noticed this one Artist kept popping up with amazing regularity, i.e daily. Eventually, I looked through history and saw I had played 300+ hours of this one album. I deleted those from history, and the recommendations for that artist dropped to what I would normally expect.
I guess I should test this. I’ll take a group I never listen to in a genre I never listen to, and set one album on repeat playback when I am not listening to the system normally (sleep, work, out of the house activities), and see how long it takes before Roon starts including the group or genre in the recommendations and chart it out by playtime.
I think that I’m rather picky with music albums. From dozens and dozens of albums in genres I like I will
probalby choose one or two maybe. The other thing is that I listen music which is a bit less popular so there is less albums with such a music. Combining these two factors may be the reason why I’m not so happy with Recommendations and New Release in Roon. That’s why I was wondering if Roon will learn in time what is my music taste and will start giving more accurate recommendations.
Other problem with Roon which I noticed may be incomplete or too general genre mathching for an albums. I already have seen albums with only one general genre like Pop/Rock without additional sub-genre description. I don’t have knowledge how recommendations works, but album described with such a simple genre can’t be recommended properly if I listen more specific music. Although it may be the one I could like.
I am similar, but I think it’s a bit too much to expect Roon to know exactly what I would like because I don’t even know myself which one or two out of dozens I will like. All that Roon can do is to recommend the ones where it makes sense for me to look closer, and to avoid those that I certainly will not like, and this seems to work quite well.
I would guess that it takes genres into account to some extent but I would think that much of it will be statistics like “people who listened to A also listened to B” which doesn’t rely on static metadata like genres
I wonder if you might have more success if you deleted your listening history (if possible) and played specific “favourites” for a few days which might improve your results. Backup everything first if you decide to try it.
I play a lot of material from Qobuz (most of which i don’t keep as a favourite in my library) which makes “new for you” pretty much useless in my case.
Otherwise this all makes sense for me as well and is noticeably different to yours:
