Gordon Haverland pisze: > On November 20, 2009, drew Roberts wrote: >> On Thu, Nov 19, 2009 at 9:06 PM, Gordon Haverland >> >>> We can observe correlations between people's complete >>> libraries, which I believe (with no proof) is what Last.fm >>> does. We could look at the time series of songs that people >>> listen to: what is the probability that a person wants to >>> listen to classical music within N different songs? >> If we want to do something like this right, will we not need >> some player cooperation? A toggle to send along info as to >> whether this series of plays is from a personally hand ordered >> playlist or not. (personally chosen order versus randome, >> computer chosen or "other" chosen of some sort. >> >> snip > > I think it would be useful to know whether the scrobbling is > coming from a hand ordered list, from some music player's idea of > random, or from some external source. Supposedly a hand compiled > list what not see the user skipping to the end of a song. For > using some music player's definition of random, you might find > users more often skipping to the end of song. In listening to an > external source, the only thing which comes to mind is the > Love/Hate tags which Last.fm uses (or used to use). > > Do any music players give information on music sources? How many > have Love/Hate options? >
Doesn't the playing order is enough information in each case? User-made playlists are of course the best source of data, but in case of lists randomly generated by player, user usually skips song he doesn't like, and playlists from external source (internet radio) are generated by some person (dj?) using some rules, which give us some sort of similarity between them. And also there are reasons why user has selected this station among lots of others, which makes dj-selected tracks somehow similar to this user's tracks. But I have to admit, if we had an information about the souce of playlist, it would be useful. But not every player has that information and it's not a part of last.fm API (which libre.fm implements). Love/hate would be great. I think we can use it if player supports it. And when we do, more players will support it :-) > >> It's great that you spent some time on sharing your >> thoughts though. I wish you success in finding great job as >> soon as possible. > > Me too. I'm starting to notice that people are finding more > obtuse than usual, which is not a good thing. I don't like to be > seen that way. > My dictionary says "obtuse = stupid, dull", which makes me a little confused about the meaning of these sentences :-S I hope you don't think I find you obtuse! > >> Doesn't just random mean random? What do you exactly have >> on your mind? > > Some people think of shuffling a deck of cards when playing > songs at random. One idea of random is we shuffle the deck > (well). We then draw songs in order until the deck is exhausted. > In that version, there is no probability of playing the same song > more than once, unless that song occurs in the person's > collection more than once. In that situation, it is possible that > the same song can be played multiple times, one after another. > > Another option is drawing cards at random, with replacement. We > draw a card (song) and start playing it. We then put the card > back in the deck, and reshuffle. This mechanism never ends (the > deck is never exhausted). Again, we have the probability of > listening to the same song back to back, with one different song > in between, .... > > We can incorporate nearest neighbour correlations, if song A > plays, it makes it unlikely that B plays next and it makes it > likely that C plays next. This can get carried to higher and > higher order correlations. > > There are some albums which tell a story across all the songs of > the album. Imaginos by Blue Oyster Cult comes to mind. If in the > course of drawing songs at random, what happens if I draw a song > from that album? Do I want to play the entire album in order? Do > I want to play from that song to the end of album in order? Do I > want to just play that song? What happens if the next draw (the > draw after, the draw after that, ...) is from that same album? > > As I said, there are lots of things involved in playing songs "at > random" that people can take issue with. > Ok :-) Now I get it. You are talking about the process of generating lists of recommended tracks and I was thinking about just about defining what is similar to what. Indeed, when we connect these two processes, we get recommendation engine ;-) I think at the beginning we should keep things as simple as possible, and later add options. Therefore I have a sugestion - let's just get some (for example a half) most recommended tracks and play them in random order (not putting a card back to the deck). Same thing could be done for albums - we could of course play top recommended albums in random order, when user would like to hear their stories. Maybe we could have a few algorithms to coose from, configured per user? > But if you listen to user fortran2's radio station > (currently at 347446 plays), you are going to find jumps because I > listen to classical, country, rock, folk, jazz, blues and some > other stuff. > Yes, and these jumps are just what I'd like to happen when I listen to your radio station. I expect it to be the mix of all the things you listen to. BTW 347446 plays... impressive. I hope you transferred them to libre.fm ;-) >>> To consider text, some people are interested in >>> constructing "random text". Perhaps a good example for >>> more complicated text construction is Lorem Ipsum. > >> I don't get what text has to do with music. Or is it some >> sort of parallell? > > Building things at random is all inherently the same process. It > doesn't matter whether you are making random integers, random > float points that are uniform 0-1, or anything else. But > applications such as random text and random music have > expectations (correlations) that people may want to use. But > working with extended correlations does come with a computational > burden. You need to draw more than one random number to find the > next song. But time consuming part is comparing the other random > numbers to complicated probability functions in order to decide > yes or no (or to pick a song). > Is it really that complicated? Maybe some example(s) will help? I think if we stored the similarities in database, instead of computing them for each song played, it could be quite simple. >> Do you mean when somebody listens to a classical track, >> he'll probably want to listen to another one? If so, that's >> sort of I was thinking also. > > Sorry, I'm getting obtuse here. If someone is looking for > recommendations on rock, the fact I listen to stuff other than > rock shouldn't preclude my listening habits from being useful. > right. > If what goes into recommendations is only the songs themselves, > this is easily done. If a person is looking at correlations > between songs (what is the probability of hearing Don't Fear (The > Reaper) immediatley after listening to Sharp Dressed Man), then > people whose listening habits cross gendres can introduce > problems. Immediately after isn't as much a problem as one (or > more) song in between. Because the person looking for 1 song > inbetween correlations, isn't expecting The Four Seasons by > Vivaldi as the song in between. > I'll have to think that over. It's a little late now ;-) Perhaps some "real life example" would help me :-) I think Vivaldi between two Alice Cooper tracks is not a problem, as long as you are listening to a radio station of some person who likes them all, or to a radio station of Alice Cooper, which 70% of fans like also Vivaldi (which I'm not sure is the case). >>> The above is handwaving theory. If we look at music >>> statistics from the music business over the last N years, >>> we run into problems. If you listen to TOP-40, you >>> will (almost?) never run across a country/western song. >>> If you look at music sales statistics, there is an >>> overlap between Top-40 and Country/Western. > >> What kind of overlap are you thinking about? That people who >> buy top-40 buy also country and western? > > Probably being obtuse again. There are false statistics out there > as well. And the split between Country/Western and Top-40 is a > good example. I think there are a number of people who listen to > Top-40 and hate Country/Western because they have been > conditioned to think that way. And I think there are a number of > people who started listening to Country/Western because they > wanted to be different (not following the masses by listening to > Top-40). Should recommendations try to bridge the gap? > I think it should. You always can skip the song if you don't like it, right? I think the possibility that top40 and c/w listeners have something similar in their tastes (for example they all like also classical music) makes it possible to fill the gap, because there will be some indirect similarity. >> But also it's very possible >> that neither top-40 listeners nor country/western listeners >> will come across libre.fm ;-) > > Actually, I think that the "better DJs" (at radio stations) out > there would probably do well to have their playlists entered as > users at Libre.fm. And then they would be able to get > recommendations of other songs to play, instead of just pushing > whatever the music people want pushed that week. :-) Or just play libre.fm in radio ;-) > I'll have to think about your second note a bit Maciej. I just hope I havent't messed it up with my English. > > Take care. > Gord > _______________________________________________ > Libre-fm mailing list > [email protected] > http://lists.autonomo.us/mailman/listinfo/libre-fm _______________________________________________ Libre-fm mailing list [email protected] http://lists.autonomo.us/mailman/listinfo/libre-fm
