Deciphering Google sets

The last few days, I’ve been messing around with an application over in Google Labs, called Google Sets. What it does is look at a short list of search terms you’ve entered and spit back a list of its own predictions as to what else belongs in that set. It’s harder to explain than to demonstrate.

The example they give is this:

mercedes benz

Submitted, those three search terms bring back the following list of items:

mercedes benz

Okay, a list of automobile manufacturers. I can understand that.

I’ve had some odd results and/or lack of results when I’ve tried doing queries there. I posted some observations in a feedback e-mail to the Google Labs group, hoping someone might tell me if I’m just misunderstanding how Sets works or if I’m finding flaws in it. It is in Google Labs, after all, their testing ground for new applications.

After the jump, I’m including a copy of my post below since it might help others who have similar questions.

Greetings! I’ve been trying some different sets out tonight, but Google isn’t on the same wavelength as I am. Maybe I’m not understanding its purpose correctly. Maybe you could tell me if I’m just not using the tool correctly or if I am misinterpreting its methodology? I love your applications and am very curious — I’d truly like to know! Here are examples for you.


First, an example which netted no results:

tongue depressor

I was thinking of things related to a doctor’s office or medical supplies.


The following didn’t bring up any results, either:



This is an instance of a good result in terms of number, but lacking in relevance (to me at least). On it, I entered these criteria:

You’ve Got Mail
Sleepless in Seattle
When Harry Met Sally

I was thinking of movies starring Meg Ryan. I’m not sure what Google was thinking! This is the predictions I got back:

sleepless in seattle
when harry met sally
you’ve got mail
pretty woman
love actually
the notebook
notting hill
steel magnolias
harry connick jr.

I guess it was going on the basis of movies with romantic plots. I’d say chick flicks, but Casablanca doesn’t fit that mold. I’m not sure why Harry Connick Jr popped up in there. He certainly wasn’t in all of those movies.

What would be great is if you guys could do something like Pandora has with Music Genome Project, providing a glimpse into why something comes up in the results. If you click on a button asking Pandora why it’s playing a given selection, you might get an answer such as, “We’re playing this song because you’ve indicated a liking for music with a strong melody line, a syncopated beat and piano accompaniment.” See what I mean?

Thanks for letting me bend your ear. Have a great Wednesday!


Gardenwife’s Plot

If I receive a reply, I’ll post it at the end of this entry. In the meantime, why don’t you head over to Google Sets and try a few of your own. Leave a comment about any fun or confusing responses you get, or do a blog entry and leave a comment with a link to it. I’m curious to see what others experience with this feature.

Leave a Reply

Your email address will not be published. Required fields are marked *