Re: Multi-categorization? (I.e. crossposting?)

[email protected] said:
> Has anyone experimented with filing messages into multiple classes
> simultaneously?  What would be reasonable criteria?  Jason, do you
> have a sense of how large the distance "typically" is between the
> scores of "good" and "bad" matches? 

Btw, in case you are interested, this problem (often called "multi-label 
classification") is just beginning to be seriously explored in the text 
classification research community.  I've already seen one very 
interesting approach in the literature:

  Ueda, N. and Saito, K., "Singleshot detection of multi-category text using
  parametric mixture models," ACM SIG Knowledge Discovery and Data Mining
  (SIGKDD2002), pp. 626-631, 2002.

I'm also planning on doing some work on this problem for my thesis.

Jason D. M. Rennie
MIT AI Lab
[email protected]
(617) 253-5339
http://www.ai.mit.edu/~jrennie/