I have serious doubts about software that could automatically determine whether an article is interesting or not, and would settle for a way to suppress duplicates as a meme propagates through the blogosphere, e.g. by finding posts that link to the same URLs.

I did implement a simplistic kill-file style filtering system, that works reasonably well, no false positives and 204 out of 1701 items in the last 7 days caught.

It’s not a Bayesean filter like SpamBayes, but it does “guess which new headlines you’re most likely to read.” It is “automatic filtering of items and recommendations based on past behavior.” You don’t have to manually construct filters or “tell the computer what you want.” Findory Blogory will “ride shotgun, watch and learn.”

Take a look. I’d love to hear your thoughts.

This isn’t handwashing OCD behavior here - I don’t read or even linger on more than a small fraction of what ends up on the screen. It’s a lot of quick glancing, page-down, and mass delete. It’s pretty much as much time as I used to spend just between a half-dozen sites like slashdot.org and cnet.com, only now my sources are so much more varied.

The challenge is that wanting to squeeze more coverage of more sources into my daily browsing has me wanting to learn more about certain technologies. That, and just dealing with more volumes of data is kinda fun in a geeky sort of way.

And, no, I don’t record all the channels on the TV. Besides watching 1 or 2 shows, and playing a few video games, this is what I do instead of sitting in front of the tube.

Oh well you learn something new everyday.

It’s a general-purpose Bayesian classifier.

Scenario #1: You Google for topics of interest. The hits returned can’t be reliably filtered by creation date, and they consider (in general) the entire indexed Web. It’s a good thing, if scatter-shot. Queries can only be refined so much.

Scenario #2: You amass and manage a list of syndicated feeds in which you have a passing interest. The feed content is indexed (Bloglines does this; Google does or will, too). You can query this universe discretely, using Google-like devices and maybe a few new ones.

What’s different about these two?

In #2, you selected (and casually manage) the universe. Its chunks — though perhaps truncated or paraphrased by the author — are reliably date-stamped. A free-text query of such data would be very different from an identical Google query.

Add a few more discriminators and it gets better. Suppose you could discriminate by

1. Paragraph length. That might get you recent topical hits of terse or essay length, as you choose.

2. Link-text density. Filter in or out those link-collection posts.

3. Query-term density or semantic weight. Is the topic being addressed directly or simply mentioned?

4. Item weight or image density. Helps filter noisy or glitzy corporate feeds (think Ziff-Davis).

There are plenty more.

The main differences, though, are date discrimination and the pre-filter of a confined universe. Me, I’m down to thirty feeds and ignore half of those. Better tools might change that.

LQ

Here are two very interesting posts on the question of Feed Readers: Glassdog asks “What’s wrong with Feed Readers” Rasterweb talks about “More aggregator maddness” Info Freako This is of more than passing interest to me of course, as it relates to the…

Here are two very interesting posts on the question of Feed Readers: Glassdog asks “What’s wrong with Feed Readers” Rasterweb talks about “More aggregator maddness” Info Freako This is of more than passing interest to me of course, as it relates to the…

Whether a blog feed is in RSS or Atom doesn’t really matter. What’s important is what we do with all this information.

Les Orchard, yes the one with the twitch for more info, argues that he is past arguing about syndication formats? I dunno, Les, call me a silly rabbit, somehow I am unable to envision Mark Pilgrim, nevermind Dave Winer,…