deeje @ BloggerJack: “Russell Beattie: Personalized Search ideas
Russell’s got a post about putting a registration-driven, personalized front end on the Google search API to build a service that takes your profile data–akin to FOAF–and then makes your search more ‘relevant’ in contextualizing it. In addition, the se [Susan Mernit’s Blog: Navigating the Info Jungle]
What if it took my weblog and personalized my search results based on that? Then I wouldn’t need to fill out yet another registration form, and it woudl continue to be personalized as my interests morph over time :-)”
It’s interesting to see the initial struggles in identifying the “right” way to personalize search results.
It’s important to recognize that the driver behind this (excluding search advertising economics), should be focused first on relevance, and perhaps second on coverage.
While I’m fully open to using collaborative filtering + social networking as a potential indicator of my interests, it seems like the initial feedback on Eurekster (social network collaborative front-end to search launched a couple weeks back) will hold for some time… that is, that group affinity and size will have a substantial impact on the currently implemented algorithms.
Let’s use Deeje and I (whom one would assume would be good proxies given that we’ve been friends for a decade and are both deeply involved in tech) as a quick example. Deeje’s searches on “Java”, being more skewed to the development side, would likely focus on the development language or coffee (i.e., development fuel), whereas mine would lean toward vacation islands. In other words, social group affinity is not a clear filter for ambiguity.
Even the above doesn’t hold. My daytime searches /would/ probably be about java development, web app servers, etc… whereas my night time searches might be vacation oriented. No wait, my daytime *weekday* searches would be development, whereas weekend searches… well, you get the point.
Determining a user’s state of mind is a difficult thing. Blogs, registration data, purchasing data, social networks… all possibly important components, but none are *the* answer.