Comments on: Elusive Information http://www.helenemartin.com/2010-06-elusive-information/?utm_source=rss&utm_medium=rss&utm_campaign=elusive-information You don't understand something until you've taught a teenager to teach a computer to do it. Wed, 25 Jan 2012 03:01:53 +0000 hourly 1 http://wordpress.org/?v=3.2.1 By: Hélène Martin http://www.helenemartin.com/2010-06-elusive-information/comment-page-1/#comment-242 Hélène Martin Sun, 20 Jun 2010 17:42:43 +0000 http://www.helenemartin.com/?p=337#comment-242 Insightful points. You're absolutely right that there's a very real tradeoff between speed and usefulness. I'd be willing to wait five minutes for truly excellent results but that would be a suicide move on Google's part. I agree that my first query was overly ambiguous... except Google "sees" me search for class materials day after day and saves my search history. It'd be nice if some of that could come into play. Similarly, it knows when I've clicked on a link and immediately left it or if I've been shown a particular result before... then again, maybe I just think using that information would make my experience better but it actually wouldn't. I'm sure people much smarter than me have been thinking about these problems for a long time! Google did have the up/down arrows for a bit which would have been great if only it weren't so easy to abuse. Insightful points. You’re absolutely right that there’s a very real tradeoff between speed and usefulness. I’d be willing to wait five minutes for truly excellent results but that would be a suicide move on Google’s part.

I agree that my first query was overly ambiguous… except Google “sees” me search for class materials day after day and saves my search history. It’d be nice if some of that could come into play. Similarly, it knows when I’ve clicked on a link and immediately left it or if I’ve been shown a particular result before… then again, maybe I just think using that information would make my experience better but it actually wouldn’t. I’m sure people much smarter than me have been thinking about these problems for a long time! Google did have the up/down arrows for a bit which would have been great if only it weren’t so easy to abuse.

]]> By: Clint Tseng http://www.helenemartin.com/2010-06-elusive-information/comment-page-1/#comment-240 Clint Tseng Sun, 20 Jun 2010 17:12:08 +0000 http://www.helenemartin.com/?p=337#comment-240 The problem here is not so much that search engines and their writers don't know how to make them more intelligent, it's that they don't know how to do so in an efficient manner. Mapreduce works well for finding keywords and creating a basic graph of what keywords point to what and in what order, because that's all static information that can be very quickly precalculated. Holistically parsing a document for meaning is an entirely different process that doesn't lend itself nearly as well to scale and parallelization. And, on the other side of the pipe, 3-8 word queries generally don't give enough context to search engines for them to figure out which of two possible meanings you want. Even a human wouldn't necessarily be able to extract your intent out of your first query. Sadly, RDF(a) won't really solve this problem, as it's really just about more keywords. (PS "while loop exercises" seems to yield decent results) The problem here is not so much that search engines and their writers don’t know how to make them more intelligent, it’s that they don’t know how to do so in an efficient manner. Mapreduce works well for finding keywords and creating a basic graph of what keywords point to what and in what order, because that’s all static information that can be very quickly precalculated. Holistically parsing a document for meaning is an entirely different process that doesn’t lend itself nearly as well to scale and parallelization.

And, on the other side of the pipe, 3-8 word queries generally don’t give enough context to search engines for them to figure out which of two possible meanings you want. Even a human wouldn’t necessarily be able to extract your intent out of your first query.

Sadly, RDF(a) won’t really solve this problem, as it’s really just about more keywords.

(PS “while loop exercises” seems to yield decent results)

]]>