I want to be pleasantly surprised by software that notices when message patterns indicate the formation of a group or project, and volunteers to set up folders and filters for me. Likewise, I want to be pleasantly surprised by an RSS newsreader that notices how I save and organize items from my subscribed feeds. No breakthrough in artificial intelligence is needed to make this happen. We do the pattern recognition ourselves, quite naturally, as we process our information flows. If software paid more attention to what we attend to, and how, there could be more pleasant surprises. Full story at [InfoWorld.com].
I had a bit of a pleasant surprise today. Last night Greg Reinacker wrote to say that Outlook 2003 does have a Bayesian filtering capability (as I'd heard), and he's getting pretty good mileage out of it, though he admits there's no documentation on whether or how to train it, and points out that it's odd there's a "Not Junk" button but no "Junk This" button. For about a week I'd been using Outlook 2003 in parallel with Outlook 2000 + SpamBayes. In OL2003 I kept manually dragging spams -- mostly Sobig.F's -- into the junk folder, but it didn't seem to catch on. Nor did it catch more than a few non-Sobig.F's. Then, last night, I loaded up a batch of about 5000 choice spam messages from my SpamBayes training database into OL2003's junk folder. That seemed to do the trick. Now it's doing much better at catching non-Sobig.F's, though for reasons I can't determine it still hasn't trained on Sobig.F's, even though they're decorated with SpamAssassin headers. Is it possible it only looks at the body, not the header? Anyway, it's not as lame as I thought, though not nearly -- so far as I can tell -- as useful as SpamBayes.
Former URL: http://weblog.infoworld.com/udell/2003/08/30.html#a788