Repository navigation
Replies: 4 comments
its telling which artists are related to each other. usually albums in common or worked on tracks together. its that simple and it's just cool view that is mostly useless except for aesthetics lmao
it was a side effect of animations code provided in #1139 (constellation map), so this is what looked useful using this ui, nothing deep lmao as i said it's extra and i would have kept it hidden but didn't cuz "why not"
how would listens decide they are related? no its just from metadata
personally wouldn't think its useful. this map just feels like relationship viewer, not smth useful for genres/styles |
I saw that after more use. That's pretty cool and I really think there is room to give it some content so it's actually useful and not just gorgeous. In fact when I saw it was with another person using the app together and both thought 'oh I can see related artists, finally I can just create a mix of this artist and related ones'. So it was a natural conclusion to expect that as soon as we saw it.
Well you already have quite a few places where you are recommending tracks based on things listened along. So I thought that was your design. My fault, Regarding current functionality I would add some actions to such page so it's a complete feature. For ex. you already show artists which have albums/tracks in common, that's a pretty great start to show "similar artists". Not based on listens nor last.fm similar data or things like that. Not perfect, but in fact that's a similar artists map in some way (and maybe this can be refined in the future). So we can zoom in and check a node (artist) and see related ones: Clicking on it opens the artist page, that's ok for now. But long clicking on it also opens the artist popup 
Since all such options are already found at the regular artist page, they are not needed at all here. What I would add? Essentially new options to play/see tracks by all "related" artists. Just imagine how cool it would be creating a mix of Artist A and all related artists. Possible options for such popup:
Play next | Play last (applied to related) No edit tags, advanced, etc. Not needed at all since it would just be a "discovery" tool. Additional cool options:
Alt-J here is related to PJ Sin suela, but such artist is also related to Aura Bae. 
As far as I understand related links, options above would be set by default to immediate links (1 node/link distance). Therefore clicking on Alt-J to create mixes, it would recommend all those blue nodes up to PJ Sin Suela. But allowing to tweak it, you could allow multiple values, so recommendations can extend not only to immediately related artists but also to artists related to the latter, etc. Essentially extending the mix to the upper pink nodes too: I think that would be extremely cool as a discovery/recommendation tool. A lot of hip-hop artists have collaborations between them and share similar sounding tracks. Other possibilities:
|
My point is you already have a framework for the animations/nodes etc. If you read my suggestions above, you can use the relations viewer to create mixes based on it. Now translated into genres/styles, we are talking about a different map which simply displays relationships between genres/styles. Not based on tracks data but a pre-generated list (you can simply hide all genre/styles not present on user library). The looks and functionality would reuse 100% the same code you already have, but aim would be a pretty different one. Instead of creating mixes from Artist A and related ones, it could be used to see relations between user-tracks on library based on genre/styles AND create mixes by genre/style similarity. And we are not talking about simply matching a value. i.e. if Track A has "Rock" genre then create a mix with more "Rock" tracks. But real similarity: I like Hyperpop, then create a mix with Hyperpop, Digicore, Glitchcore, ... That's a huge change regarding recommendations since you are not matching anymore specific values, just looking for similar sounding/related things. For ex. this is my implementation in foobar, without a fancy animated map. I just ask it to play similar styles to 'Cloud Rap'. All offline. Library statistics for a given genre: (obviously this kind of UI is terrible xd) Select one track and create a mix based on a genre/style similarity algorithm: (the track has multiple genres/styles and we are not simply matching all tracks with the same values; also note how the mix created is sorted with harmonic mixing rules using Key tag -key column in color-) Obviously I'm not talking about adding all this into Namida, just that a simplified design can be added using the same map feature for genres/styles. Where we could look at all the genres/styles of our library, using ralations/links based on the github repository I linked, and create mixes with it. Most streaming services have recommendation mixes and similarity based recommendations, so this could cover such features while being 100% offline and not having any kind of bias regarding cluster of listens. Personally I would find it pretty useful. Having the power to create mixes of similar tracks without needing complex smart playlists, a trillion of listens, online data, machine learning or too heavy processing would be awesome. I do it all the time on my PC, I want to listen to rolling stones and similar things and just click on the button and bam a complete playlist which "makes sense", while not being hardcoded to just "same genre", and allowing me to also listen to "new" things or things which are not strictly the same but are "similar". It's like putting a thematic radio, it just plays somewhat related tracks which sound good together but with real variation. EDIT: I would obviously be willing to help in whatever is needed regarding this part, the link creation algorithm, etc. |
|
They are 2 different features, but having any of them would greatly improve Namida as a recommendation service. Note this is something usually not found on any player (apart from the usual "smart mixes" like "your top tracks") and reaching Spotify-like quality level- There are also like thousands of requests/webs/threads on the net regarding this kind of features: https://spotalike.com/en Being able to do this without spotify, third-party services, offline, no AI and just on your phone would be a pretty great feature to highlight. And since Namida already supports youtube I can imagine adding the genre similarity part could also be used along it to even recommend new tracks -not found on library- using Youtube, so it opens further possibilities there... |
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
I have seen multiple iterations to the artist discover feature (which is a bit hidden until you notice it by chance!), and I'm interested on understanding its purpose.
I see the first page shows this:
Then I also noticed this one:
My impression is it's telling us artists which are similar (?) based on listening history I guess.
Could you share your ideas about it? I suppose you have at least some design in your mind regarding its future purpose. Also it would help to focus feedback or giving you possible ideas about it.
So far I have only find it interesting from a visual POV since clicking on things only creates playlists/queues for selected artists (not the cluster). Is that right?
Also I think I found an issue with the current animations at 7.4.0. Note how focusing on a node, moves 3 or 4 columns at right (instead of the focused or immediate nodes). Not sure if that's intended, looks weird to only see the animation at right.
Finally I would love to see something similar for genres/styles, and using it for genre-similarity based recommendations. I think I already mentioned it in an issue some time ago, but limited to a genre map, it would be relatively easy to adapt this: https://github.com/regorxxx/Music-Graph (at least the visual map, with your framework instead of my horrible html)
Within namida, now that you have the same for artists, is just a matter of clustering based on similarity instead of listens. Would you be interesting on something like that?
All reactions