Google Sets

So not even close

[Sebastian]

We all know about one of Google's beta tools - Google sets ( [labs.google.com...] ). I've tracked many of Google former Beta tools, like news, and have found them to be immensely accurate, and not remotely betaish (add this word to your dictionaries). They've been simply amazing.

However, not so with Google sets. For the terms I've used in the past, the results have been SO bad.

Firstly, you create a long list of words to form a topic or set. Then click on the button. The results? Well, it basically returns the exact same words I gave it originally. Well, that didn't take much work for it. Then, it suggests some completely ridiculous words that aren't even close. I'm sorry, but Google Sets receives an F.

Don't believe me? Let's form an arbitary set, as an on-the-spot example. We'll do this together:

We give it the terms: cat, pussy, alley cat, tom cat, tabby.

See the pattern yet?

Google's response: Cat, Pussy.

Excuse me, but they're the terms I gave you in the first place! How much programming did it take you to figure that out?

Now, granted, it's only in beta stage, but considering every other Google beta I've tried has been nothing short of amazing, I feel Google sets (as it is now) is a total failure.

[ciml]

You get good results if you try words that tend to be found together on Web pages and are very commonly used.

The results for orange and banana compare interestingly with the results for orange and blue, IMO.

[vitaplease]

Sebastian,

also check the comments here: [webmasterworld.com...]

[dcheney]

It seems to do fairly well with things like a series of presidents, heads of state, dog breeds, etc. But doesn't do as well with more obscure things.

[vitaplease]

there are some observations and comments on:

[groups.google.com...]

Google seems to use the whole Google index:

as the mentioned set for "blonde" and "brunette" shows "safe search" seams disabled.

Looking at results it does really seem to be mainly webindex based, as many results of the proposed sets show captial letters for the first character, where they would appear as first word in for example titles.

[Sebastian]

If Google is planning on using Google Sets as the underlying technology for one day factoring in theming into calculating PR, I think SEOs have very little fear of theming being introduced in Google soon.

However, Google may well introduce theming using an entirely different assumption engine.

[vitaplease]

they probably use stuff described in here for their sets.

­http://www.inma.ucl.ac.be/~blondel/publications/02BS-syn.pdf

can anyone get the http: //skeptic.stanford.edu/data/ online thesaurus mentioned in that paper to work?

[rfgdxm1]

>You get good results if you try words that tend to be found together on Web pages and are very commonly used.

Not necessarily so. As a test I entered a couple of slang terms for a not at all common recreational drug. Google sets came up with 3 other slang terms and the name of the drug itself. Thus Google sets also can work well for no so common situations.

[victor]

When it's good, it's impressive.

But when it misses the mark, it goes flying off into who knows where.

I tried Tinker Tailor Soldier and asked for a large set.

Google sets had just one word for me ....

Camper

Bad AI -- no donut :)

[dkoller]

I don't think it is very fair to give google sets an 'F', just because it did not work for one of the sets you provided.

I played around with google sets quite a bit, and most of the time it gave very relevant results, and in the few instances where it wasn't great I could easily modify a term or two in my set to get more relevant results.

This is a great tool for research, writing, etc. It's like a big thesaurus that extends beyond words to include people, places, and ideas.

I would give google sets a 'B+' in its current form. It could be much better, but after all it is in beta. The technology is sound, I believe, and who else is doing it?