If it’s not, it’s still pretty interesting. Philipp gives a detailed overview of a forum posting from a fellow who claims to be an engineer deep in the bowels of Google. Great stuff in here. Including:
“The Google application process is annoyingly slow and can easily take months sometimes. Anyone who’s really interested in working there would be much better off finding an employee to refer them”
“Nobody keeps track of 20% time with any care whatsoever. It’s assumed that, if a deadline is pressing on your main project, you’ll work on that.”
“…you can divide internet traffic into five approximate and unequal segments: porn, spam, corporate, knowledge, and personal.”
I am not entirely sure that his claim that Google News came out 20% think tank is accurate. Alltheweb had the News Search implementation prior to Google News. Google News was a response to alltheweb news.(if webmasterworld.com keeps archives search for alltheweb oneups google).
“… you can divide internet traffic into five approximate and unequal segments …”
Philipp later took this one back, saying he’d mistaken it. Very interesting all the same. A few other noteworthy bits:
– his Gmail account storage limit is 1 TB.
– “After being asked what it takes to get fired at Google, Zorba replies that abusing logins is a fireable offense.” Especially if you get anywhere near the log system. Which makes sense … the “database of intentions” is that log, and there’s a *lot* of information there that Google likes having, doesn’t want getting out, and doesn’t want insiders abusing.
– Zorba says it was depressing to be able to use Google Maps [before it was launched], but not be able to print out routes to take them along as that might breach confidentiality. (I’d agree, and frankly, I’d think if it’s only a few weeks away, what if a few people see it. so what?)
And the anecdote about Larry and Sergey with a laptop + webcam strapped to a remote-control jeep is funny.
Yes, I know of published academic work that described GoogleNews-like systems, at least 2-3 years before Google News. There is probably more work that I am not aware of, that is even older. So if Google News did come out of that 20% time, they must have spent that 20% time reading conference papers…
Surely Google News isn’t that complicated. I haven’t read any papers on it, but having looked at it, it’s fairly obvious what they’re doing.
They crawl news sites (or just subscribe to their RSS feeds) and cluster (http://www.elet.polimi.it/upload/matteucc/Clustering/tutorial_html/) the stories they find, hence identifying which ones are talking about the same thing. You can take advantage of the fact that most news sites will classify each story into a section (Google probably maps each site’s section breakdown to their own) so you can group the different story clusters that way too.
Of course the implementation will be non-trivial, and there will be fine tuning, but the concept is pretty simple.