On Tue, 9 Nov 1999, Mayhem & Chaos Coordinator wrote:
> Any progress on this yet? Are you still considering it?
Current status: I'd still love to, but can't make it double as
a research project. :) So, I probably won't start on it until
I've gotten that school project out of the way...
Feel free to discuss design, though.. :) I'm not so sure
about using a neural net now (I'm pretty clueless about how to
implement one at the moment), but it should work almost as well
to just work out a scoring system, sort songs by score, and
randomly select a song which is within roughly 1 standard
deviation from the top of the list. The scoring factors were
listed in an earlier message:
> > - "karma" points (user's musical preference)
> > - file access dates (older is better)
> > - file creation dates (newer is better)
> > - time of day (slow down at night?)
> > - relative song position (song X sounds good after song Y, but
> > clashes with song Z, etc...)
> > - play history (karma points over time may average into a
> > better value than raw points)
If possible, I might try to create a system which can be
generalized to other types of playlists, to arrange any type of
viewable data better than "random". Maybe not.
Another thing which needs to be designed is a good interface.
Any ideas?
_ _ _ _ ___ ___ ------------"Use the source, Luke!"---------
( \/ ( \/ (__ (__ ) | Scott Scriven (Toy Keeper / XYZZ) |
\ / \ / // // | mailto:[EMAIL PROTECTED] |
/ \ / / //_ //_ | irc:serdevian.dyn.omnipotent.net |
(_/\_(_/ (___(___) | http://www.vis.colostate.edu/~scriven/ |
