h a l f b a k e r y
May contain nuts.
add, search, annotate, link, view, overview, recent, by name, random
news, help, about, links, report a problem
or get an account
Take any search engine company that wants to improve. In their list of search results to any query, they should offer check boxes with the ability for the user to rate each result . The choices are: "Useful", "Useful, but not to me", "Useless", and "No comment".
Based on the checked off boxes,
the search engine can analyze the contents of the "Useless" pages and learn which keywords are often present on pages that are positive matches to the given query string, but which no users really want. Future queries on those same search terms will bring up more relevant pages.
Here's the beauty part - only users who do a search and fill out the ratings for the top 20 links every so often - say once a week - get the improved search engine. Don't help us improve and you just get the standard engine.
You enter your ratings (subject to a human challenge question to make sure it's not an automated spam robot rating its own pages very highly). Click SUBMIT, and you are then taken to a new version of your search query and hopefully you will see improvement in the number of relevant links and the absence of previously useless "false" links.
Note: Google (and maybe others) are trying to improve their image search engines by inviting users to play a free game where an image is shown to two different users and if the users agree on a keyword, then the keyword is added to the image as a tag. Such a system is deficient. It's deficient because even the blurriest photo of a bird will be duly marked by both users as "bird" and will henceforth show up on all bird related queries, to the chagrin of future searchers. Search engines could possibly analyze the user's behavior in scanning results (which result links did they skip over? which ones did they click on? Did they return to the search listing again soon after?) but this is slow and greatly inaccurate.
Asking the user directly would improve the analysis.
[hidden truths, Dec 28 2007]
Great talk on ESP game by sp: Luis von Ahn
uses human computation, which is proposed in the idea. [4whom, Jan 01 2008]
||There will, however, be many people who try to poison the results, simply because they are low-functioning trolls with nothing more fulfilling in their lives. Whether or not this works will depend on how significant this effect is.
||Google has some things in the pipeline like this, I've heard. I think it involves removing select results and having similar results removed with them.
||I always assumed that search results are prioritised by the number of people who clicked each link, is this not the case? If so, then this is kind of baked.
||//Note: Google (and maybe others) are trying to improve their image search engines by inviting users to play a free game where an image is shown to two different users and if the users agree on a keyword, then the keyword is added to the image as a tag.//
Brought to you by the same brain that founded CAPTCHAs, Louis van Aght. You may have a different opinion of his work if you listened to his methodologies. Will post link.