7 May 2005

Tags and emergent systems

I had previously lamented the limitations of a semantically empty tagging system (e.g. de.lici.us), and suggested that incorporation of the wonderful WordNet would solve that. The same tag system would be used, but a search tool would be added that allows ontological searches (e.g. a search "flower" would match the "Hydrangea" tag).

Before that post, I also rambled about tagging as a subset of centralized emergent systems on the Web (e.g. flickr or MySpace) as opposed to distributed systems (I had suggested harnessing existing Web sites to create more specific tools, a la Paul Rademacher's stunning use of Google Maps to automatically map Craig's List apartment listings).

The benefit of an emergent system is that you only need to define the rules of the system, not the content. AI long ago discovered the brittleness of creating top-down systems (e.g. write every rule of logic) compared to the fuzzy elegance of bottom-up systems (have genetic algorithms discover those logic rules). Often, these two approaches are merged to harness the benefits of both. Tags seem to be a dumb form of bottom-up system. This could be merged with WordNet (or something similar). I originally felt that the deficiencies of tagging were fixable-but-flawed. Based on a project I'm about to start, I'm now beginning to think that emergent systems are the best method for creating Web content that will evolve. Rather than being "brittle" ontologies (as Tim Bray suggested), I think tags should be viewed more as emergent ontologies.

[ posted by sstrader on 7 May 2005 at 12:00:14 PM in Programming ]