Explaining My Search Idea Further
As a follow-up to my earlier post, I thought I’d try and explain what I’m thinking in terms of personal search.
The idea is based on Paul Graham’s "A Plan for Spam" which uses Bayesian Logic on emails to determine which mails for you are spam, and which are not. The general principle is that if you mark an email as "spam", it adds all words in the mail to a total for words being spam. Any mail left unmarked marks all the words in it as non-spam. Then, each time a mail arrives, it is checked for whether, on balance, the contents are spam or otherwise.
I’m not looking for spam, just search results that don’t match with what you want. I’m wondering if the same principle would work.
So, if I search for "Prince", I get pages covering everyone from the pop performer, the project management methodology, the tennis company, the golf course in Sandwich and Prince Charles. I’m not too interested in golf, tennis or the heir to the throne, but I’m quite interested in project management and pop geniuses.
So, I click the pages for those I don’t want, and a load of words get marked down. The pages I leave get marked up. So, words like Charles, golf, tennis, championship get marked down. And words like Purple, Kiss, Partyman and Methodology get marked up. So, next time I search, these get considered, and over time, tuned to what I want.
Over time, the search gets more "intelligent" and improves around what the user searches for.
It could even be that just playing around with such an idea could yield some interesting results.
RSS
Just to spin this further, you’ll need a way to make the search "forget" your previous preferences as well.
Let’s say so far you’ve been interested in the singer and project management. Now you suddenly need to research the heir to the throne for whatever reason. Based on your previous searches/interests those results typically get filtered out, making it more difficult for you to find what you need instead of making it easier.
Somehow you’ll need to address this little problem to make the search really effective…
BTW, I’m not 100% sure, but I think when you search using a Google account Google at least tries to use a way similar to what you’re suggesting. But I might be completely wrong here. I’ve just got the funny feeling that I’ve read this somewhere.