Re: Effect of widely differing volumes on ifile classification

Jason Rennie <[email protected]>:

> Possibly.  We could limit the database to (e.g.) 5,000 word
> occurrences per folder.  When a new message exceeds the limit, we
> scale the (aggregate) word counts down and round.  It'd be similar
> to decaying the old word counts with each new message.  I haven't
> tried this, though.

if you do this, you have the counts of other folders present.  can
you do something about the "skew" at the same time?  btw, is there an
URL to that paper you mentioned?

  clemens