Re: [Rhythmbox-devel] automatic rating and weighted random

[Paul Kuliniewicz <kuliniew purdue edu>]

On Mon, Mar 29, 2004 at 03:00:14PM +0200, Benjamin Otte wrote:
> On Fri, 26 Mar 2004, Paul Kuliniewicz wrote:
> 
> > I'd say it should only autorate in response to an explicit action by the
> > user.  If the user skips, decrease the rating.  If the user starts
> > playing a particular song, increase the rating.  But when the algorithm
> > decides what to play next, leave the ratings alone.
> >
> That would mean that for me (play whole library with shuffle, press next
> when I don't like a song) the algorithm would only rate down and I would
> end up with 2000 songs rated as 0.
> So that idea doesn't sound very good to me.

Good point.  I was thinking that if you don't change what Rhythmbox
decides to play, it did a good job figuring out your tastes, and there
was no need to modify ratings.  That clearly doesn't work if you don't
jump *to* anything in particular, though.

Looking at the modifiers used for autorating, it looks like a song's
rating stabilizes if you skip it once out of every five plays.  So even
if you only skip a song once in a while its rating won't go all the way
up to 5.

Getting back to the original question, since the rating modifier for not
skipping a song that Rhythmbox picked is pretty small, a
autorating/play-by-rating feedback loop's effects wouldn't be very
noticable.  It's probably not worth worrying about unless there's
evidence that a problem really exists.

Digital signature