Personally, I think Bayesian filtering is interesting, but I’ve been trying off and on to find something a little more effective…

Aggregator + Bayesian Filter

Each time you read a post you can mark it as interesting or not. The filter has two buckets, appropriately, the interesting bucket and the uninteresting bucket. Interesting posts should have certain similar features, usually in the form of buzzwords I would guess, which the filter should be reasonably good at picking out. And the interesting vs non-interesting sorting capabilities would act as a way of making the “foam” float to the top. And then one should be able to feel totally unguilty on those busy days when you hit “Mark Uninteresting As Read.”

When I worked for my college paper, I loved to “surf” the AP newswire (on our minicomputer terminals!). Around that time, I read “The Media Lab” by Stewart Brand, who pointed out that even a big newspaper prints no more than about 10 percent of the AP stories on their wire.

Then, maybe 10 percent of what gets into the paper actually catches the reader’s eye. That means that 1 percent of the potential information flow gets to the average reader, and generally, it’s not the right 1 percent.

When I worked for CNN.com, we tried to create a custom news product (called, strangely enough, CustomNews, launched around 1999) that let the reader twiddle a few dials to tune the news that got your attention. You could, for instance, tell the application what sports teams you followed, and set up individualized keywords (I might choose “Apple”, “photography”, “Honda”, “Atlanta”, “Tucker, GA”), and stories featuring those keywords would come up in your news.

CNN had a big advantage — we could rely on metadata included in the newswire. When you’re pulling newsfeeds from the web, you have a much richer tapestry, but you don’t know much about the threads.

The River of News approach looks to concentrate on maximizing flow, but what we really need, as you suggest, is a way to find the fish in the River of News.

I rely on some keyword feeds from Google News, Feedster, and Blogdigger to get a similar end result, but none of them matches up to a human editor. It’s getting to the point, for instance, that I know I can skip surfing Mac news, and rely on John Gruber at Daring Fireball to catch almost anything I would be interested in.

And there can be a quite a backlog in the lower-priority folders. So I wish there was a way to just “show the latest 50 entries across all the feeds in this folder” so I could read what’s fresh without having to read or throw away the backlog.

I’ve done this with regular, big M, media at 180n.com. I added user voting on top stories to make it bit more interesting, but you see the idea. I call it WebRSS.