Semantic taxonomy, part 2

Last month on 28 April, I wrote an entry about semantic taxomony. In my post I referenced two abstracts I found on the web via Google searches: both academic exercises on the topic of semantic taxonomies as they relate to personalized web searching.

The first abstract was authored by disciples of business and computer science. This paper proposed that users should anticipate their search needs and preferences and "teach" the system what kind of results they want. While I have always been a proponent of search preferences and offering options to the user for searching, including selections that can be stored for later use, I do not advocate making the user guess what kind of results they think they want and forcing them to teach the search engine to find them. This might work for a small population of search experts who comprehend the inner workings of the engine's algorithm(s) but for the vast majority of users, it would be a befuddling experience.

The second abstract was authored by adherents of linguistics and computer science. This study championed an intelligent system that observes the user's searching and based on compiled empirical data, the engine builds a profile and anticipates the user's preferences. It refers to the data in this profile as "meta-knowledge." I am a big fan of this approach. I believe it serves the majority of individuals with search needs much better than laborious form-driven exercises that may or may not aid in finding anything and have ephemeral relevance.

CNN recently published an article on a new spate of search engines (naming Twine, Hakia, SearchMe, Kosmix, Cuil, Duck Duck Go, Scoopler, and Wolfram Alpha) that have built this kind of meta-knowledge into their engines' algorithms and are aggregating them againt results from sites such as Twitter. These new engines start where Google currently ends, observing the user's behavior and providing better and better results over time.

I imagine once one or two of these engines breaks ahead of the pack, they will be absorbed into the Google omniplex. This will be a welcome addition to the Google toolset -- right now Google seems to only use our histories to show us tailored advertising. Though I prefer an ad that offers something I might want over one simply based on keyword matching against my search query I would really enjoy this sort of intuiting within the results themselves.