[mythtv] Bayesian analysis

[Kirby Vandivort]

On Sat, Nov 15, 2003 at 04:09:00PM +0000, Steve Hill wrote:
> On Sat, 15 Nov 2003, Kirby Vandivort wrote:
> 
> > Well, mine give me 500 programs that its not recording that I've never
> > heard of.  :)
> 
> :)
> What sort of scores did it give the programs?  The top ones on my list are 
> all things that I record, and score very highly (e.g. 98% or more).

http://www.ks.uiuc.edu/~kvandivo/bayestest.html

> > At any rate, we need to get together on this.  I'm currently in the
> > process of writing some code that will add rankings to the user interface
> > where users can rate random programs that they like or dislike and the
> > step i was planning after that was to do the bayesian.  (perhaps you've
> > looked at the recommendations patches that I sent to the list a couple
> > of months ago.  If not, I would recommend trying it out)
> 
> I hadn't spotted them - I'll have to have a look.  I'm not convinced that 
> treating all the programs that don't get watched as "bad" is a valid idea 
> - they should probably be neutral.  By reading the rankings then there is 

neutral is what I was planning.  -5 to 5 is what I had figured, internally,
with anything that isn't manually changed getting a zero value.

> a valid source of bad programs so the main listings can be neutral.  Maybe 
> a good method is to present the top 10 programs the bayesian analysis 
> finds to the user and let them rate them, that way it might weed out the 
> false positives.
> 
> One of my thoughts (which I mentioned in the readme) is that it might be 
> good to have a central database of viewing habits (for those people who 
> don't have a problem with submitting that kind of data).  Because with a 
> reasonably sized database you could probably extract amazon.com-style 
> "People who watched 'Startrek' also watched 'Firefly'" type details, which 
> may be more beneficial than self-analysis.

Yep...  collab filtering is something else that has been discussed.

http://www.gossamer-threads.com/perl/mailarc/gforum.cgi?post=80673;search_string=recommendations;guest=1571452&t=search_engine#80673

has some data and links in it that would be useful for you to peruse.

-- 

Kirby Vandivort                      Theoretical and 
Senior Research Programmer            Computational Biophysics 
Email: kvandivo at ks.uiuc.edu          3051 Beckman Institute
http://www.ks.uiuc.edu/~kvandivo/    University of Illinois
Phone: (217) 244-5711                405 N. Mathews Ave
Fax  : (217) 244-6078                Urbana, IL  61801, USA