Grokker: The Slow Search

May 19, 2005 4:19 AM

Grokker. It's powered by Yahoo! search, but the results are presented in a very different way, a visual map [example]. There's a few tools to refine the output. It's different, a bit slower than a "normal" search and requires a bit of patience. There's an affiliated weblog, with an entry explaining their philosophy, "Moving Beyond the Algorithm."
posted by gsb (31 comments total)

Kills my FireFox everytime.
posted by homodigitalis at 4:29 AM on May 19, 2005

Almost forgot, a counterpoint.
posted by gsb at 4:30 AM on May 19, 2005

If this means that my students will be familiar with the word 'grok' again, I'm all for it.
However, in Firefox, it claims I don't have Java.
In Safari, it ran, and it looks neat, but felt a little clunky. I could imagine using something like this to rapidly get acquainted with the landscape of an unfamiliar topic. But when I did a search on "differential geometry" I found it very hard to locate significant things I knew should be there. Also - just my stupidity - it took me a minute to figure out how to actually get to the sites.

Finally, since it's inevitable...
posted by Wolfdog at 4:31 AM on May 19, 2005

Doesn't work in my ff either. I'm not sure of the efficiency of having to load up JVM every time I want to run a search. Still, interesting idea.
posted by blag at 4:54 AM on May 19, 2005

looks interesting but the visual representation is non-intuitive and operates by some obfuscated, byzantine logic which actually impedes understanding. It sould be outlawed.
posted by clockzero at 5:05 AM on May 19, 2005

My test search was "Fnord." I found it pretty cool, especially how it separated out all the Robert Anton Wilson related pages. There my be some advanced efficiency in this idea.

However, I hated how my cursor covered the mouseover information.
posted by sourwookie at 5:33 AM on May 19, 2005

It works on my Firefox, but I agree with the blogger at gsb's counterpoint link: too many clicks, not enough intuition in the blobs.

However, I don't always agree with this: "A simple list of text links really isn’t a broken format, and you can’t help wondering what the graphical designers are trying to fix." The list of text links is great if your result appears in the top 3 or 5 or 10, but it is clumsy for searches that return huges results. This seems to be the main advantage of a graphical interface (see Grokker's blog for an illustration of a "Peeps" search), but I don't find Grokker all that helpful in solving the problem.
posted by Gordon Smith at 5:43 AM on May 19, 2005

I'm not sure of the efficiency of having to load up JVM every time I want to run a search

I'm so non-thrilled with Java that it wasn't even worth it for me to fire it up just to try this once. If I already have an operating system running, I really see no need to load another one on top of it just to do something silly. I really, truly have yet to see a single good reason for Java on a website. Java is almost always (mis)used on websites for some effect that could be accomplished much more easily (from an end-user standpoint) using CSS or flash. While this Grokker search clearly goes beyond what CSS can do, how difficult would it be to use Flash instead, like the news aggregrator does?

It's not necessarily a bad idea, but the implementation of it seems to blow.
posted by caution live frogs at 6:10 AM on May 19, 2005

i think the people at grokker need to grok what a search engine is supposed to do: work quickly, weed out chaff, not give me results so biased by advertising that they are useless (see google) - and let me do the thinking myself.

also anyone who doesn't follow the simple, no clutter, fast loading front page of google is missing the boat.
posted by three blind mice at 6:17 AM on May 19, 2005

What exactly is the difference between firing up the java plugin vs firing up the flash plugin froggo?
posted by zeoslap at 6:17 AM on May 19, 2005

It makes Firefox choke, and IE go out with an exception. Great program, guys!

Spock does not grok this.
posted by Harry at 6:49 AM on May 19, 2005

Worked fine on my instance of Firefox, was kinda crap, but it certainly worked.
posted by zeoslap at 6:59 AM on May 19, 2005

If you go here, you'll end up at Ken Perlin's page, and see the exact Javascript that the interface uses, albeit an erlier version.

And there's absolutely no mention of it on the Grokker site. How very rude.
posted by dirtynumbangelboy at 7:24 AM on May 19, 2005

Crikey! That's pretty bad/stupid of them.

An aside, I should have checked it out in other browsers. I'm using Safari2, Firefox gives me the "you have no Java" message. Obviously buggy as hell.

And the on the Java side, it's probably one of the fastest applets I've tried on OSX -- compared to Touchgraphs and the like.
posted by gsb at 7:33 AM on May 19, 2005

It works just fine in Mozilla 1.7.7 with Sun JRE version 1.4.2_08 on Windows 2000 SP4. I'll have to test it on my VectorLinux box later.

As an aside.....this is damn cool.
posted by PROD_TPSL at 7:42 AM on May 19, 2005

From "moving beyond the algorithm:" The algorithm is no longer relevant.

Seriously. I wonder if Knuth has been told yet.
posted by nervousfritz at 7:49 AM on May 19, 2005

Works in my Firefox, but I can't see the use in it. I searched on a subject I know a bit about, and got a screen full of dots. Hovering my cursor over the various dots is not nearly as easy as scanning text looking for a key word or phrase, to me.

Feels to me like an example of trying to fix what ain't broke.
posted by BoringPostcards at 8:13 AM on May 19, 2005

Runs well on a Safari 2/2x2GHZ G5. Whether it is *worth* running is a moot question. . . .
posted by rdone at 8:43 AM on May 19, 2005

Works fine in IE and Netscape, here. Personally, I don't think the wait for results is excessive and you only have to load up the virtual machine the first time that it is used while your browser is open. Closing it down and starting up a new window, however, loaded up another instance of the virtual machine... So I guess it has a few bugs. Just a note, dirtynumbangelboy, the applet is compiled java code, not javascript which, while having similar syntax, is completely different. Just had it pounded into my head in a web programming class.

Did you notice the tools at the bottom of the page? It allows you to search within your map, enter a cutoff date or yahoo search rank. And if the circles annoy you, you can change them to squares. A better use of real estate in my opinion. I love the categorization of results. Seems easier to find things in a search that returns hundreds of seemingly random results...
posted by Roger Dodger at 8:50 AM on May 19, 2005

It seems to me the advantage in this sort of search would be that you could separate out whole classes of information that occur with the same search term. For example, there's a toy piano band that goes by the name Twink. As some of you may know, that's a rather unfortunate name when you're doing a search, as there's a huge class of search results that has nothing to do with toy piano music. When I search for "twink" in Grokker, I get nine circles, including "Toy Piano Music", "CD", and "Gay". I know right away that the "Twink" I'm looking for won't be under "Gay". So I check out the other circles instead.

Now, granted, this simple example probably doesn't provide a clear advantage over just scrolling past unrelated results in Google. Nor is it necessarily much better than just scrolling past uninteresting results in Google. However, I can see that this sort of approach (not necessarily Grokker's particular implementation) could make it easier to weed out completely unrelated ideas that occur with similar search terms.
posted by afiler at 9:10 AM on May 19, 2005

Erm, the second sentence in that last paragraph should be:

Nor in this case is it necessarily better than just adding on another term, like "twink toy piano music".
posted by afiler at 9:12 AM on May 19, 2005

The project I am involved with has looked at Grokker as a working example of the sort of thing we want to do for our browsing interface. I didn't realize they had made it an applet, so thanks for the linky. It does what it does well, but as others have mentioned, it is sort of trying to solve a problem that doesn't exist.

Visual finding aids don't fare so well in a keyword search environment because users tend to be faster/more accurate evaluators of relevance than the algorithms that power these interfaces, though they can be useful in the case of homonyms. OTOH, a visual finding aid is incredibly useful when engaging in browsing behaviors. If done properly, a visual representation can efficiently give information about relationships, proximity, and frequency. Grokker does the first alright and the third to some degree but in order to be really useful it should also spatially represent the semantic proximity between clusters. To do this on the fly is really effing hard though. But with a finite collection and a fixed hierarchical structure, this all becomes quite a bit easier and faster. Or at least this is what we're finding in the course of our project. IMO this would be a great thing to put on top of DMOZ, LCSH, or--as it appears in this case--Yahoo!'s categories as an alternate browsing interface.

Oh, and JRE1.5+FF1.0.4+XP works great. Will try on Gentoo box at home tonight.
posted by Fezboy! at 9:20 AM on May 19, 2005

Hrm, I'm a bit more fond of something like teoma which suggests additional related search parameters, although I'm not exactly convinced that Teoma's method really helps that much. What would be nice would be a "disambiguation" sidebar. So for example, searching on "star wars" would provide a disambiguation hint for "movies" and "strategic defense initiative."

Doesn't seem to work well with my browser and jvm, but I have a slow system.
posted by KirkJobSluder at 10:28 AM on May 19, 2005

Grokker's clustering is really valuable. Clustering can help you weed out a whole category of things you're not looking for, yes, but it can do a lot more than that. It can also point you to things you didn't know before, connections you may not have realized.

Something like Google takes a blinders approach. It teaches you to master the engine's arcane syntax and use it to restrict your query as much as possible. It's good for finding something you know you're looking for; it's not for exploring. That's where clustering comes in. Clustering encourages you to throw out really wide-ranging queries and actually explore the subject.

Unfortunately, Grokker's interface--while it does have a significant initial "wow" factor--just isn't terribly useful. Rivadeneira & Bederson showed that Vivisimo's textual clusters actually are much more useful, and now Clusty is much more friendly to the average web surfer. AOL picked up Vivisimo's clustering for their web search, and Microsoft's been running some experiments with the idea, too.

In the end, Grokker is very neat, and the clustering idea is really revolutionary. But its implementation is just too unusable, so it's damned it to never being much more than a toy.
posted by jefgodesky at 11:45 AM on May 19, 2005

Seems like vivisimo's search clustering with musicplasma's presentation.
posted by jungturk at 12:01 PM on May 19, 2005

The presentation is ancient ... goes back to the Windows Explorer interface, really. Vivisimo's been doing this since 2002. It looks like musicplasma started this year....

(Though, I can't get musicplasma to actually respond....)
posted by jefgodesky at 12:47 PM on May 19, 2005

Ohhhhhhh ... OK.

(Just got musicplasma to actually do something....)

Yeah, jungturk, that's pretty much it.
posted by jefgodesky at 1:07 PM on May 19, 2005

What's next? GrokWhack?

[yeah, I know, it's not a true whack 'cos it's not in the dictionary but I ran out of patience and my other whack didn't work (even though it did in Yahoo)]
posted by carbon at 9:11 PM on May 19, 2005

wtf is a whack?
posted by puke & cry at 9:29 PM on May 19, 2005

ah ok
posted by puke & cry at 9:31 PM on May 19, 2005

zeoslap - Flash doesn't bog my system down while it loads up a new OS on top of my existing OS. Once that Java thing loads it sits there, memory-resident, until I close my browser. Sort of like a web page opening WINE to run a Windows app, or having to emulate Classic while running OSX. Flash is much faster and makes use of the OS I already have installed.

The Java plug-in just sucks so much juice I normally browse with it deactivated. As I said, the time it takes to load is really not worth the end results, at least 90% of the time I've seen it used. (No, thank you, I do not wish to load Java just to punch a monkey in your banner ad. Yes, scrolling menu options are neat-looking, but jesus is it worth the load time? And so on.)
posted by caution live frogs at 6:59 AM on May 20, 2005