Gnus 5.0 and Grouplens

Gnus 5.0 and Grouplens

[Brad Miller]

>>>>> "Lance" == Lance A Brown <labrown@dg-rtp.dg.com> writes:

Lance> Hi,
Lance> Is there anyone else out there useing Grouplens in Gnus 5.0 or
Lance> September Gnus?  If so I'd like to talk to you about how you are using
Lance> Grouplens and what you think of the contributions I've made.  I keep
Lance> diddling around with displaying the prediction value in different
Lance> ways.
Well, I am using september gnus.

I would be interested in your observations on how the GroupLens
predictions match up against the scores you get from the gnus score
files. 

Right now, I think the nicest feature of GroupLens gnus is that I can
read groups using the ',' key to take me to the next highest scored
article. 

What I would like to figure out how to do is adjust the scoring somehow
so I can expunge the articles with low predictions < 2 in rec.humor for
example, and see the articles with high predictions plus I'd like to see
all of the articles for which I have no prediction.  The trouble is that
my default score is 0.  Is there any way to tell the expunger to throw
out articles below the expunge threshold unless (= score default-score)?


Lance> Also, when can we expect to see some confidence numbers in the BBB
Lance> output?

As soon as we figure out how to calculate some confidence numbers that
are meaningful. :-)  So far, most of my efforts have been spent on just
getting the whole BBB system stabilized (we seem to be there now).  Over
the next few weeks I plan on spending a lot of time working on and
analyzing the algorithms.  

We've collected a LOT of ratings 30,000+ over the last six weeks, so
I've got some good data to work with in trying to improve the current
prediction algorithms, and hopefully get some confidence numbers.

\Brad
----------------------------------------------------------------------------
Brad Miller                    | e-mail: bmiller@cs.umn.edu
University of Minnesota        | phone:  (612) 626-8396     
Department of Computer Science | www:    http://www.cs.umn.edu/~bmiller
EE/CS 5-244                    | 
----------------------------------------------------------------------------
**  Save USENET Try GroupLens: http://www.cs.umn.edu/Research/GroupLens   **
----------------------------------------------------------------------------

Re: Gnus 5.0 and Grouplens

[Lance A. Brown]

Brad Miller <bmiller@cs.umn.edu> writes:

> I would be interested in your observations on how the GroupLens
> predictions match up against the scores you get from the gnus score
> files. 

I don't see much correlation right now, but don't think this indicates
anything wrong.  I use the Gnu scoring mechanism to do 2 things
mainly:

1.  Use adapative scoring to push threads I am currently reading to
    the top of the Summary buffer

2.  Use large negative temp scores to force spam and off-topic threads
    to be expunged so I dont' even see them. :-)

I use Grouplens in non-override mode so that I can see both my normal
Gnus score and the Grouplens prediction and find that I am more likely
to pick a low-scored article when it has a high Grouplens prediction
and have found some useful and/or enjoyable threads that way.

> Right now, I think the nicest feature of GroupLens gnus is that I can
> read groups using the ',' key to take me to the next highest scored
> article. 

I don't see enough predictions in the groups I read (mostly
comp.os.linux.*) to be able to do this.  Normally only 0% - 10% of the
articles have predictions.

> What I would like to figure out how to do is adjust the scoring somehow
> so I can expunge the articles with low predictions < 2 in rec.humor for
> example, and see the articles with high predictions plus I'd like to see
> all of the articles for which I have no prediction.  The trouble is that
> my default score is 0.  Is there any way to tell the expunger to throw
> out articles below the expunge threshold unless (= score default-score)?

I think this would require another gnus-*-score-by-* function
that can adjust the Gnus score according to its Grouplens prediction
value.

--[Lance]