David Sifry has been having second thoughts over the Easy News Topic self-categorization of weblog RSS feeds; some of his qualms are people mis-classifying themselves, others in semantics that don't lexically match (synonyms). As evidence, David offers the failure of Meta tags such as keyword.

I have no evidence one way or the other, but I remain more hopefull in the power of unfettered anarchy to evolve its own solutions. What I see on the existing self-placement sites, which includes both the simple hotlist indexes that let you place your link in some box just as the new trackback blog-directories like TopicExchange, what I see is everyone first rush to stake their own category, and the taxonomy goes flat. But then new subscribers come in and realize no one is going to find them at the bottom of a long list of 1-link categories, so they choose an existing bucket that seems likely to be travelled ... and pretty soon we see 80% rush to get in the 20% top categories.

But here again, people are driven by two competing objectives, one to be uniquely differentiated, the other to be found in a group, so as these top topics grow to unmanageable size (themselves flatten out), people will fork them, recursively repeating the top-level problem, eventually moving into a deeper hierarchy they see as still likely to attract those who got as far as that hot topic just above them. In the long run, chaos will anneal to a solution.

Ditto for the synonym problem: If two directories have lexically different topics that are similar, and both directories grow in popularity, it won't take long before people learn of the other and double-classify their blog into both taxonomies ... and by being present in both, they inadvertently create the needed bubble bridge our spider robots can use to suss out the semantic relationship.

Why didn't this happen with the Meta tags? Simple: There was no feedback. No vehicles grew up to take advantage of any of those meta tags (except description being used for bookmarks) and as a result (except for description) no one really took them seriously. If my keywords had been structured the way RSS 1.0 taxo: namespace intended, and if these were then mined by the crawlers so it meant automatic entry into correct branches at Yahoo and DMOZ &c, ie instant gratification for the effort, I have no doubt people would have taken it more seriously. ENT will be our test of my theory.

Since my definition of 'Knowledge Management' jibes with the M-word used in 'Forest Management' or 'Wildlife Management', maybe we can't control or mandate this growth, but we can foster it through our expert seeding of a few categories that fit wide generalizations; this is the approach taken with Yahoo, DMOZ and others who constrain what they accept into categories that are only forked by administrative control.

To control the size of the searchspace for directory users, we could flatten the recommended ENT into the depth of heirarchy where we're comfortable, perhaps even allow these deeper recommended structures to appear as soon as they gain sufficient member counts. This would also mean that if people had chosen a popular taxonomy on some other website and then 20 of their members submitted to our site, by the weight of their numbers, they'd automatically create the corresponding taxonomy in our hierarchy. Instant standardization based not on abstraction, but generalization of actual data!

Whether we actually have the science to make intelligent predictions that would better what might naturally occur is a question up for debate, and for that reason, I'm hoping at least one of these self-classifying directory servers gives us the control group by staying their course with the chaotic anarchist evolution.