Moved the blog from Radio on my old OS 9.2 iBook to a new OS X Titanium. What a bloody mess trying to get the right files, and only the right files, moved over, and still losing half the config information. Bleah! It's enough to make you switch to Moveable Type...

Permalinks seem to be alive once more after disappearing during the transition for a while. I'm now running Safari and seeing the bug reported by Kevin Kelly, where the cute little calendar floats in the center of the page instead of staying left. Any CSS wizards out who can look at the source and tell me why it's working OK in IE and Netscape and wandering around in Safari?

Update: Tip of the hat to Brian Bechtel at Apple for telling me I was encountering a Safari rendering issue that is fixed in the latest beta release. If anyone else is seeing the same problem, go get a new download. Thanks, Brian!
9:04:28 PM    

Having recently written about Power Laws and then Latent Semantics, I got to thinking last night. One problem with those cute power law graphs is they're lumping together everyone from Alterman to Oxblog to Dave Winer and Gizmodo. One wonders if they are really competing with one another, and these interesting link clusterings at Ross Mayfield's blog suggest that in many cases there can be nearly disjoint populations of readers. See where I'm going?

The first pass experiment would be to run some vanilla clustering algorithms, minimum spanning trees perhaps, over link graphs from the blogosphere. Then do power law charts on the extracted clusters separately. The cycle budget for that one shouldn't be too bad.

Next move is to take the resulting clusters, and do latent semantics (PCA) on a window of recent posts in each blog to characterize the leading factors describing each cluster, and by the way make a very nice conceptual search over them.

Now for extra credit, and a cycle burn that will make Intel cheer, do the PCA up front, over as much of the blogosphere as you can handle. One trick we pulled in the Guides project (PDF) mentioned below was to augment the feature vectors for each document with the presence or absence of a link to each other document (blog in this case). Yes, that does indeed cause the feature matrix to grow as the square of the number of blogs indexed, but fortunately it's very sparse so it can probably be dealt with, modulo lots of memory and cycles. Representing the links may get around the potential problem of conflating Alterman and Josh Chafetz just because they use similar words, with very different audiences. Now you've got a completely factored index over the entire blogosphere. Rerun as desired until your CPUs melt, and you get a dynamic view of the shifting topic and blog population patterns. Just think, you lurking .gov types, a top level synopsis of the 'Internet street' delivered to you daily! Dual use, sell it to Google! What say?

11:08:14 AM    

Just read this.
10:03:18 AM