Google’s Udi Manber – Search is a Hard Problem from Read/Write Web has an interesting write up on a presentation that Udi Manber, Google’s VP of Engineering, gave the other day at Supernova. In his presentation, he covered why “search is a hard problem.” But a few of his statements really stood out from the others.
Manber said 20 to 25% of the queries that Google sees today have never seen before. Wow, 25% of the queries people enter into Google today, were not seen before today?
Manber also said that Google is rolling out a new search function that interprets questions better. Manber gave three examples of how Google will be handling this.
(1) “How much does it cost for an exhaust system” will pull up results from “cost of an exhaust system.” (2) “Overhead view of bellagio pool” will pull up results from “bellagio pool pictures.” (3) “Fedora 5 losing network connections” will pull up results from “fedora 5 network configuration.”
Postscript From Danny: Google promised this query expansion last month as part of its Searchology event, and Google’s Matt Cutts told me this weekend that it is already live in the service.
Postscript From Barry: That 25% New Queries Figure? Ballpark Estimate, Says Google is the update we just got from Google.
Related Topics: Google: Web Search | Search Features: Query Refinement | Stats: Search Behavior
I blogged about the 25% of new queries issue at http://www.blogstorm.co.uk/blog/mahalo-human-powered-search/
If 25% are new then the database of previously searched for terms is growing by about 1.5 billion every single month – on reason human powered search can never keep up.
Interesting, I just didn’t think that number was so high.
Heh… maybe that “queries that have never been seen before” number is gonna see a little uptick following that recent Googlenope story – I know I entered some garbage searches to try to trick The Google myself!
It’s definitely that high. The long tail is very, very long.
The 25% figure has been announced previously. It came out in Google’s 2006 press day for sure (Danny blogged about it on SEW) and it was mentioned in the Searchology Press Day this year. I think the other points were also raised by Udi at Searchology.
Matt, it seems like you’re confirming 6 billion queries a month at Google. If so, the various search engine ranking reports released by HitWise, Nielsen, and comScore are seriously underestimating Google’s activity.
I coined the term “the invisible long tail” to explain the phenomenon in a Youmoz article a couple of weeks ago. There, I explain a technique I use to identify such terms.
Identity followed up with a clever strategy to predict such terms through patterns.
I have heard of many people using Google as a spellchecker. Could This be where they are coming from?
We’ve spent some time analyzing the AOL data release of late last year and have arrived at similar conclusions. These are described in a series of blog posts but in essence:
- The top 1% of search phrases = one-third of search volume
- The top 25% of search phrases = two-thirds of search volume
- The remaining 75% of search phrases are sought only once in the three month period for which data was released.
In the Sunday,8/5/2007,Washington Post Magazine, Gene Weingarten had a follow-up column to his original Googlenopes column. He mentions a new website, http://googlenope.com, which is a reader contest on these. It’s pretty clever; definitely worth checking out.