Google Suggest

Cristiano writes "As you type into the search box, Google Suggest guesses what you're typing and offers suggestions in real time. This is similar to Google's 'Did you mean?' feature that offers alternative spellings for your query after you search, except that it works in real time." It crashes Konqueror, but works nicely on Mozilla. Update: 12/11 by J : The engineer who thought of it, then built it in his "20% time," blogs about the process.

Comments

Cool! Just like form AutoComplete

[GillBates0]

but with a HUGE database/archive of possible candidates at it's disposal.

Wonder how it'll hold up when it gets out Beta though...it's bound to be pretty computationally intensive.

Re:Cool! Just like form AutoComplete

[TheRaven64]

Of course, if you're trying to decide between two search algorithms using the same dataset and one algorithm is O(n log n) and the other is O(n^2) then the Big-Oh comparison actually means something.

Not always. Big-Oh only means the worst case. Quicksort is O(n^2), but the worst case is so rare that it usually performs better than many O(n log n) algorithms.

Re:Cool! Just like form AutoComplete

[JabberWokky]

If it makes you feel any better about Slashdot, there are plenty of people who knew exactly what you're asking and haven't responded because they know enough to know that they don't know the answer. Thus you're only getting responses from people who didn't understand the question.

I'd guess that n will vary between now and when they release as they grow their database.

Interestingly, they seem to clamp down on search phrases that are synonyms and start with the same beginning. For instance, the search for "Rocky Horror" is more common than "Rocky Horror Picture Show", but only the latter is listed. In this case, for reasons that are specific to the search, using RHPS over Rocky Horror can cause problems as it eliminates "Rocky Horror Show", which is the stage production, and often swapped by the public for RHPS.

--
Evan

Re:Cool! Just like form AutoComplete

[Anonymous Coward]

Yes, that is, as they say, trivially obvious.

However, it is not trivially obvious what "n" is counting is this case. For instance, I can give you an O(1) algorithm for a search of an arbitrarily large database, as long as I'm measuring complexity by w.r.t. "the number of servers used in processing the query." Adding servers will not increase the processing time, therefore there is a constant upper bound (with respect to the number of servers being used), therefore it is a constant time (or less) algorithm.

This is a stupid and intentionally misleading example, but it should show that O(f(n)) can be an utter lie if you don't know what n is. Oranges? Turtles? Pigeons? WTF?

Re:Cool! Just like form AutoComplete

[BorgCopyeditor]

Instant access to any piece of human knowledge

I'm no Luddite, but might it not be significantly more exact to add the qualification "that can be found on the Internet"?

Advertising potential

[Anonymous Coward]

I wonder how long it will take before companies are able to pay for their 'suggestions' to show up at the top of the list.

Privacy?

[Frisky070802]

I'm surprised no one has commented on privacy yet. It looks to me like it uses past queries to suggest future ones, because as soon as I zoomed in on an unusual name, it offered a couple of bizarre queries that could only have been typos (one was a two-word query, so it wasn't simply every word indexed -- unless they know every pair of words that quickly?).

I don't mind Google knowing what I ask, but I'm not sure I want the world to see them.

What Google needs is Lexis-Nexis and Journal tabs

[F34nor]

What Google really needs is a for pay LexisNexis tab so you can find real information from real sources in real time. That and a tab that indexes full text medical and science journals. Those damn journals! I love em but I don't have hundreds of dollars a year for each Psych. journal I want to read and hate going to libraries if I just want to see what's shaking in the world of science. With full text periodicals and full test journal search Google would become a singularity of information.

Cool, but dangerous to google

[Anonymous Coward]

Much as I love this, I really do, I think Google is screwing themselfs over big time with this if it were to go live. Think about it, who has the most use for up-to-date common searches? Link farmers, google`s biggest problem!

All they have to do is, search for 'a', create links with the sugestions as text, search for 'b'... etc. Voila, a link farm optimized for the favourite searches of all google users. This can be automated to stay up-to-date. Much faster, more extensive and more acurate the googles zeitgeist.

I really hope googlecan make this it work though... its more helptfull the eclipse and zsh completion together ;-). Its also quite revealing of how much my searches are like everybody elses, or rather, arent alike at all.

Re:The ABCs of Google Complete

[kzinti]

That is based on alphabetics ordering, I think it would make more sense to sort based on rating.

Google is ranking these suggestions so that the ones you're most likely to search for are higher. So even though 's' has more hits than 'spybot', Google thinks you're more likely to search for 'spybot'. That makes sense - the terms people search for most often are not necessarily the pages with the most search results (or the highest Pagerank).

What is n?

[MasterVidBoi]

I'll try this myself: What does n represent in this case? The number of pages in google's database, the number of words or phrases in their database? The length of the search string?

I would really like to know where you came across this. Can point us to a discription of the algorithm?

Re:Google starts to suck

[That's Unpossible!]

Yes, Google is testing one of the coolest features I've ever seen on a website...

They sure are starting to suck.

There's about 3 billion women in the world...

[Anonymous Coward]

...so get to it, bucko.

Re:Obfuscated Javascript

[Jugalator]

I don't think they obfuscated it to obfuscate it, but to save bandwidth with a minimum of cost in CPU time. Take 1 byte and multiply that with however many visitors per second they're getting ;-)