Can we please get multiple titles again?

jackfrost · Post by **jackfrost** » Tue Dec 11, 2007 8:40 am

Caravan Ray wrote:OK - here is how to make Songfight! perfect.

I like that Songfight makes songwriting a sport. And the way it is set up with the no-registration, hit-a-button, first-past-the-post-is-the-winner voting is pretty well perfect as far as I'm concerned.

I also like one title - because I don't like having to choose between titles and one title sort of focus' everyone in the one direction. But the hassle of reviewing lots of entries for one title does suck a bit.

Another thing I like is the awsome big-arsed song archive that Songfight! has. Now, I don't know how possible this is cos I'm not a geek - but I reckon it would be really cool if the songs in the archive could have the good-ok-bad buttons attached to them like on Somesongs. That way - outwardly, Songfight is exactly as it always was - BUT, if you are too busy to review all the songs but you still want to give some feedback, you just wait till the fight is over and songs go to the archive, then hit a good-ok-bad button for each one, and maybe drop the odd comment or two on the boards if you feel it appropriate. In the place of lots of reviews, which may or may not be useful - I would love to have a quantifiable ranking for my songs. And the archive would take on a life of its own as old gems are rediscovered and Starfinger's More Than Soup will continue to be sung by future generations.

i'm sure if this was implimented the abuse would outweight the benefits. people would "bad" tag artists they didn't like and "good" tag themselves, along with every variation of friend flooding of the tags. one of the benefits of songfight is that all the songs are just there. there is no preferencial treatment in presenting different artists or fights. it's up to the person to wade in and see what they like or don't like. if you want someone to tell you what's "good" music then turn on the radio.

Post by jb » Tue Dec 11, 2007 9:15 am

What *I* think would be cool would be a SongFight Pandora. From Pandora's explanation of their methodology:

We believe that each individual has a unique relationship with music - no one else has tastes exactly like yours. So delivering a great radio experience to each and every listener requires an incredibly broad and deep understanding of music. That's why Pandora is based on the Music Genome Project, the most sophisticated taxonomy of musical information ever collected. It represents over seven years of analysis by our trained team of musicologists, and spans almost a century of music (and soon several centuries!).

Each song in the Music Genome Project is analyzed using up to 400 distinct musical characteristics by a trained music analyst. These attributes capture not only the musical identity of a song, but also the many significant qualities that are relevant to understanding the musical preferences of listeners. The typical music analyst working on the Music Genome Project has a four-year degree in music theory, composition or performance, has passed through a selective screening process and has completed intensive training in the Music Genome's rigorous and precise methodology. To qualify for the work, analysts must have a firm grounding in music theory, including familiarity with a wide range of styles and sounds. All analysis is done on location.

The Music Genome Project's database is built using a methodology that includes the use of precisely defined terminology, a consistent frame of reference, redundant analysis, and ongoing quality control to ensure that data integrity remains reliably high. Pandora does not use machine-listening or other forms of automated data extraction.

The Music Genome Project is updated on a continual basis with the latest releases, emerging artists, and an ever-deepening collection of catalogue titles.

By utilizing the wealth of musicological information stored in the Music Genome Project, Pandora recognizes and responds to each individual's tastes. The result is a much more personalized radio experience - stations that play music you'll love - and nothing else.

Imagine going to the SF archive and being able to play it like Pandora. Have an Octothorpe station, for example. Be able to thumbs up/down songs. Share stations with people. Etc. etc.

Heh. I love daydreaming.

JB

Rabid Garfunkel · Post by **Rabid Garfunkel** » Tue Dec 11, 2007 10:05 am

Somewhere above it was asked what the ideal fight size was. IMO, if it fits on a single audio cd, then it's golden. Anything more is, again IMO, unwieldy.

roymond · Post by **roymond** » Tue Dec 11, 2007 10:18 am

jb wrote:Imagine going to the SF archive and being able to play it like Pandora. Have an Octothorpe station, for example. Be able to thumbs up/down songs. Share stations with people. Etc. etc.

Heh. I love daydreaming.

JB

Isn't that called somesongs? Could Jefff slap that interface over the archive?

Post by jb » Tue Dec 11, 2007 10:23 am

röymond wrote:
jb wrote:Imagine going to the SF archive and being able to play it like Pandora. Have an Octothorpe station, for example. Be able to thumbs up/down songs. Share stations with people. Etc. etc.

Heh. I love daydreaming.

JB
Isn't that called somesongs? Could Jefff slap that interface over the archive?

Similar, but Pandora is less about everyone seeing what ratings a song has, and more about letting you hear things you like and exploring relationships-- but mostly hearing things you like and playing things you'll probably like, based on your selections and "thumbs".

Not saying the somesongs method couldn't be adapted, but it doesn't involve analyzing tempo, rhythm, lyrics, style, etc to find songs similar to one another. somesongs does have an agreement algorithm, but it's people-based. You could play songs that someone with 80% agreement to you likes. That wouldn't divide them into "Octothorpe" or "JBB" style divisions.

We might get close with only, say, a 50-distinction form for analyzing and some kind of db and interface to play them, with a subset of volunteers categorizing the songs, who are sworn to objectivity.

roymond · Post by **roymond** » Tue Dec 11, 2007 10:50 am

Paul Lamere is a researcher at Sun labs and does a lot of music analysis stuff for Sun, and I believe is going to be releasing some sort of service soon to allow sites to provide this kind of experience. Should be interesting.

Rabid Garfunkel · Post by **Rabid Garfunkel** » Tue Dec 11, 2007 10:53 am

The Pandora model is awesome, true, and would be übersweet implemented here. With one exception: since there's no ASCAP (or whatever it is) tracking being done at songfight.empire, please don't have a maximum number of song skips-per-hour.

Like the eskimos and snow, I'm sure a third of the distinctions would involve some variation on "crap"

I'm joking. Sort of.

fluffy · Post by **fluffy** » Tue Dec 11, 2007 11:03 am

SITM is pretty interesting, but it doesn't look like it has any sort of collaborative filter aspect to it.

Every now and then I get a twinge to try to make a community filtering engine. It's a tricky problem if you want it to scale up to a large number of users or media items, though.

Post by **Lunkhead** » Tue Dec 11, 2007 11:49 am

jb wrote:This is true, and analog being imperfect, I can't find those stats at the moment. If you want to be a pessimist, I can't stop you. Although I do think it's a lot more likely that someone would download a Front track from Front's site than from songfight.org.

I know this would involve work, and probably a significant amount of it, and it wouldn't really do much other than allow you to gather some data you probably don't even really care about, so I don't expect this to happen, but as you probably already realize, if you served the current fights' songs from a different context than the archive songs (such as "/current" instead of "/music") you could more easily differentiate in the logs the current fight traffic vs. the archive traffic...

fluffy · Post by **fluffy** » Tue Dec 11, 2007 11:57 am

Referrer logs can handle that for the most part.

Post by **Spud** » Tue Dec 11, 2007 12:00 pm

Lunkhead wrote: I know this would involve work, and probably a significant amount of it, and it wouldn't really do much other than allow you to gather some data you probably don't even really care about, so I don't expect this to happen, but as you probably already realize, if you served the current fights' songs from a different context than the archive songs (such as "/current" instead of "/music") you could more easily differentiate in the logs the current fight traffic vs. the archive traffic...

Making the change to set it up would be no particularly big deal. The only problem is that we would then have to move the files at the end of each fight. Yes, fluffy, I know that that could be automated, but now we're making more work to set THAT up...

fluffy · Post by **fluffy** » Tue Dec 11, 2007 12:14 pm

Actually I'm opposed to URLs which change. URLs should be permanent.

Post by **Lunkhead** » Tue Dec 11, 2007 12:27 pm

fluffy wrote:Referrer logs can handle that for the most part.

Good point. Theoretically, most hits to songs in current fights should be coming from the home page or from the current fights' pages, right?

fluffy wrote:Actually I'm opposed to URLs which change. URLs should be permanent.

Another good point, although, the fight URLs change when they go into the archive. Do you mean that, more specifically, file URLs should be permanent?

Post by **Spud** » Tue Dec 11, 2007 12:35 pm

The song URLs do not change, currently. The page URL does.

fluffy · Post by **fluffy** » Tue Dec 11, 2007 12:42 pm

I was referring to the file URLs. Though also to the individual fight URLs, which are also permanent.

The main page is always the current fight and in that regard it's also permanent, even though the content on that page changes. The point is that currently, URLs never expire (except in the case of a song being taken down due to a rules violation) - they might change their contents, but they never go away.

(At this point I'm very tempted to ramble on and on about idempotency and permalinks and clouds and content tagging and blah blah so hopefully you actually get what I'm saying since I don't want Frankie to yell at me again for bloating the thread or whatever.)

Post by **Lunkhead** » Tue Dec 11, 2007 12:57 pm

I get it. I responded a bit by PM. Anyway, just to get back on topic, I like multiple title fights, but I don't feel strongly about 1 vs >1.

HeuristicsInc · Post by **HeuristicsInc** » Tue Dec 11, 2007 1:14 pm

jb wrote: Not saying the somesongs method couldn't be adapted, but it doesn't involve analyzing tempo, rhythm, lyrics, style, etc to find songs similar to one another.

A Pandora-ish model would be fantastic. I love that site. This is the rub, though, and that's the big difficulty, is analyzing the songs to create the music genome. The guy that founded Pandora does these "town hall" meetings; I went to one here in Baltimore a year or so ago. Fascinating. But in his talking about how it all happens it is apparent that you need people that are objective and consistent and knowledgable about music theory to be able to correctly fill out the genome, so that's not an easy task. Actually when I was there I handed the guy a pile of HInc cds, so you can actually put me into it and hear a station of artists like me (also including me). And that's really cool, actually.

If I remember correctly the Pandora guys aren't releasing the details of the genome. We'd have to make our own, but that would be a lot of fun.
-bill

Caravan Ray · Post by **Caravan Ray** » Tue Dec 11, 2007 1:27 pm

jackfrost wrote:
Caravan Ray wrote:OK - here is how to make Songfight! perfect.

I like that Songfight makes songwriting a sport. And the way it is set up with the no-registration, hit-a-button, first-past-the-post-is-the-winner voting is pretty well perfect as far as I'm concerned.

I also like one title - because I don't like having to choose between titles and one title sort of focus' everyone in the one direction. But the hassle of reviewing lots of entries for one title does suck a bit.

Another thing I like is the awsome big-arsed song archive that Songfight! has. Now, I don't know how possible this is cos I'm not a geek - but I reckon it would be really cool if the songs in the archive could have the good-ok-bad buttons attached to them like on Somesongs. That way - outwardly, Songfight is exactly as it always was - BUT, if you are too busy to review all the songs but you still want to give some feedback, you just wait till the fight is over and songs go to the archive, then hit a good-ok-bad button for each one, and maybe drop the odd comment or two on the boards if you feel it appropriate. In the place of lots of reviews, which may or may not be useful - I would love to have a quantifiable ranking for my songs. And the archive would take on a life of its own as old gems are rediscovered and Starfinger's More Than Soup will continue to be sung by future generations.
i'm sure if this was implimented the abuse would outweight the benefits. people would "bad" tag artists they didn't like and "good" tag themselves, along with every variation of friend flooding of the tags. .

Err...yeah - that is how it would be supposed to work. It seems to work on somesongs.

fluffy · Post by **fluffy** » Tue Dec 11, 2007 1:28 pm

I don't think tagclouds are the way to go for a recommendation engine. I think for a corpus as small as Song Fight it's better to just have traditional up/down ratings and do a correlation between listeners to find recommendations.

My experience with Pandora is that a lot of its recommendations are complete junk, just because while the tags make perfect sense in the context of the song, that doesn't mean that a large overlap of tags means the songs are at all similar. (The example I always bring up is that based on a couple of Radiohead songs, Pandora likes to recommend Red Hot Chili Peppers.)

I've also done some work with 12-tone analysis stuff, and it also tends to get easily confused by the sorts of stuff I enjoy since they can't really deal with things like dissonance or complex rhythmic structures very well.

Netflix and Amazon both appear to use some sort of analog to Bayesian inference for their recommendations, and it works pretty okay once they get enough data to go on. Making such a system scale is difficult but the current size of the corpus and userbase at Song Fight is probably small enough that a naive implementation would do pretty well to start out.

jackfrost · Post by **jackfrost** » Tue Dec 11, 2007 1:55 pm

Caravan Ray wrote:
jackfrost wrote:
Caravan Ray wrote:OK - here is how to make Songfight! perfect.

I like that Songfight makes songwriting a sport. And the way it is set up with the no-registration, hit-a-button, first-past-the-post-is-the-winner voting is pretty well perfect as far as I'm concerned.

I also like one title - because I don't like having to choose between titles and one title sort of focus' everyone in the one direction. But the hassle of reviewing lots of entries for one title does suck a bit.

Another thing I like is the awsome big-arsed song archive that Songfight! has. Now, I don't know how possible this is cos I'm not a geek - but I reckon it would be really cool if the songs in the archive could have the good-ok-bad buttons attached to them like on Somesongs. That way - outwardly, Songfight is exactly as it always was - BUT, if you are too busy to review all the songs but you still want to give some feedback, you just wait till the fight is over and songs go to the archive, then hit a good-ok-bad button for each one, and maybe drop the odd comment or two on the boards if you feel it appropriate. In the place of lots of reviews, which may or may not be useful - I would love to have a quantifiable ranking for my songs. And the archive would take on a life of its own as old gems are rediscovered and Starfinger's More Than Soup will continue to be sung by future generations.
i'm sure if this was implimented the abuse would outweight the benefits. people would "bad" tag artists they didn't like and "good" tag themselves, along with every variation of friend flooding of the tags. .
Err...yeah - that is how it would be supposed to work. It seems to work on somesongs.

the way i see it: the purpose of the archive is to archive. i don't want johnny songfuck's opinion to influence what songs i would choose to listen to whenever i open up an old fight. not knowing what is behind the artist names is half the fun. why not just have a separate part of the message board for "my favorite songfight songs" and that way johnny songfuck can post (with hyperlinks) lists of his favorite artists and songs and expound on how great his taste is?

HeuristicsInc · Post by **HeuristicsInc** » Tue Dec 11, 2007 3:15 pm

fluffy wrote: My experience with Pandora is that a lot of its recommendations are complete junk, just because while the tags make perfect sense in the context of the song, that doesn't mean that a large overlap of tags means the songs are at all similar. (The example I always bring up is that based on a couple of Radiohead songs, Pandora likes to recommend Red Hot Chili Peppers.)

this is when you say "thumbs down" and tweak the results. i have found with just a few thumbs up and down for things that i get a very good correspondence with whatever i was trying to do.
-bill

j$ · Post by j$ » Tue Dec 11, 2007 5:08 pm

jackfrost wrote:i don't want johnny songfuck's opinion to influence what songs i would choose to listen to whenever i open up an old fight?

I kind of agree with this, in that Good/OK/Bad is not enough to offer any valid criticism/indication, even on a perecentage basis. Who says one can only deem a song three ways?

On the other hand, as jackfrost reveals, as soon as we start making excuses for not listening to other people's opinions, we are all 'johnny songfuck'. [cue dramatic music]