#metabrainz

hiya! I like that you're thinking in terms of UI, but currently I am much more interested in the technical details behind the semantic recommender -- that sounds like a lot of work, really. hence my question as to what tools you plan to use to make all this happen.

Hey! So most likely I feel that Python or JS with a web scraper and interaction with music brainz RDMS would allow for these connections to be identified. I say a web scraper due to the fact that sentiments can be analyzed with information that is taken off of publicly available information about albums, and then put into bins/categories

This would require a NLP approach combined with content-based recommendation

So an example could be extracting information about an album's themes/moods, analyzing mlp, and then quantifying that in the context of both musical genres, but also moods

what sites would you scrape?

"content-based recommendation" what does that mean? acoustic qualities of the music?

ah yes.

so, we have the AcousticBrainz project that has this info. but the underlying algorithms suck, so the data is... well.. shit.

that is why the project is on life support now. and spotify shut down access to their acoustic API.

so, content based recommendation is outside of our reach for the time being.

mayhem[m]: Well mainly I was looking at wikipedia due to how it contains aggregate info from different review sites. RYM is protected by cloudflare + it's against their TOS so that's not really viable lol

WP is perfect. anything else without clear usage rights would be a much harder sell.

and MB has a ton of WP links so we can download that stuff right quick.

<mayhem[m]> ""content-based recommendation..." <- Acoustic Qualities could definitely be an aspect of it, but there are also things like subgenres that better explain a particular type of music (once again, wikipedia has in depth explanations of various subgenres), as well as general descriptors that describe themes within an album, or tags that describe emotions an album evokes.

AOTY for example has detailed descriptions of genres. I know that musicbrainz does too with tags, but I feel that there could be a finer level of granularity

defining a subset of tags that express moods and using those: cake.

not sure if that data is statistically significant, however.

mayhem[m]: Which data?

tags that express moods in MB.

Ah

Yes

I'd assume the sample size is under 30 😭

however, we could encourage users to tag tracks with moods in order to improve how this works.

yeah, scraping the bottom of the barrel.

what else could be used?

have similarity data, popularity data

we have a neighborhood where users are compared to other users based on their listening history.

so you get a user's musical neighbors

mayhem[m]: I did see that which I thought was interesting

I'm just throwing out what we have, hoping something adds to the pile.

I did not understand how anything was being calculated though 😭😭😭😭😭

I have to go to bed rn because i have a long day tmrw but I'll definitely sleep on it and talk it out with a few friends!

pearson coefficient. https://en.wikipedia.org/wiki/Pearson_correlati...

yep, here too. past 1am

but lets keep chatting, this sounds super interesting

mayhem[m]: will be looking at this!!!!