{ Thursday, April 16, 2009 }
We've been busy. Since we launched Hunch, over 4.3 million questions have been answered, counting both THAY questions (Teach Hunch About You) and topic question/answers. This is phenomenal, and gives us a ton of data to work with. In the past, statistical prediction like Amazon's collaborative filtering has usually been for one subject only -- i.e. Netflix will look at the movies you like and recommend other movies. And who hasn't bought a John Grisham novel only to have Amazon recommend *other* John Grisham novels? Um, duh. Cross-topic correlations like Hunch has are likely to be more interesting.
Next week we are planning to give researchers access to our data in what we're calling the "academic" API to distinguish it from the, say, "widget" API, which will be released when the site goes public. It's all anonymous and aggregated data of course. Leave a comment below if you're a researcher who wants in.
Some remarkable correlations have emerged from the data already, some of which we posted on Twitter and on the Hunch blog:
- People who believe that alien abductions are real are more likely to blame Nancy Pelosi for the financial crisis.
- People who have broken a leg like video games such as Madden NFL 09 and NBA 2K9, whereas non-leg-breakers prefer Little Big Planet, Katamari Damacy, Super Mario Galaxy and World of Goo
- One of the best predictors of whether people agree they should switch to a Mac: whether they like to dance. Are PC users less fun? The data has spoken.
- Have you used a fake ID to do something you weren't supposed to? It's OK, we won't hold it against you here at Hunch. Fake ID users are more likely to be happy at startups.
The Facebook vs. Twitter topic brought up more interesting stuff: Facebook people, at parties "interact with many, including strangers" whereas Twitter devotees "Interact with a few, known to them". Facebookers buy shampoo that costs more than $7, Twitterers prefer to spend less on their hair, thank you very much. What else? Facebookers are significantly more likely to have had an alcoholic drink in the past 24 hours. Twitterers seem to be less social than Facebookers. But maybe it's because Twitterers report that they have oily skin -- it's the zits!
Now that we have this data, we can improve the Secret Sauce so that all of Hunch's results are tailored to not only you, but people like you who have entered in only a small amount of data (ie, logged out users). And we can add a bunch of features we've had planned but didn't have the data to implement. We can tell you more about different facets of yourself, and cluster you with other people. We can tell you how much of an outlier you are, or how non-unique. Now things get fun.
If you haven't tried Hunch yet, go sign up and we'll send you an invitation. And thanks to everyone who's already contributed.
LINK | 12:34 PM | TB
Yay! LOVE data, LOVE hunch :)
Carrie (Westlake) Cronkey | April 16, 2009 5:38 PMWOW! This news is simply awesome - 4.3 million already ... fabulous results!!
This data is fun!! Now, I think I may have missed my career calling. I should have been a Social Media Researcher! Oh wait, that wasn't an available degree back then... drat!
Hunch data set is rich with cool factoids! I see lots of different revenue models with this neat data. The potential here is really vast - nice business concept - congrats!!!
@susanbeebe
Susan Beebe | April 16, 2009 8:17 PMI would love to sign-up but apparently the system thinks my email address isn't a valid one... Hrmmmm
Jon | April 17, 2009 1:27 AMCaterina, I'm equally awed and creeped out by Hunch.
First, the promise of "tell[ing] you more about different facets of yourself" hearkens back to fortune telling and palm reading, but with real data toward an accurate description. I'll come back to that.
Second, I'm not sure I want you to "cluster [me] with other people." So much of what makes Flickr fun is what made the web fun in the first place -- following links forward to something (i.e. someone) new and wonderful. It was voluntary and based on either existing associations or (I assume) chance, much like things happen in real life. Hunch "cluster[ing]" implies a certain omniscient agency on your part; there's a deep _knowing_ in there. Wonderful and amazing a person as you are, I resist the thought of granting you that.
Third, Ms. Beebe's comment about "lots of different revenue models with this neat data" is spot-on, and it's terrifying. I think it must be said that if advertising and marketing had any interest in general human wellbeing the industries would close up shop. I doubt that the highly specific data I've given Hunch will stay anonymous -- that the wall between anonymous Hunch user and Justin the consumer isn't permeable. When you're dealing with real people with actual behaviors and true vulnerabilities, you're in moral territory. Add a business model to the mix and you're even deeper.
Now, I'm able to identify and resist marketing appeals, but certain people very close to me, whom I love and care deeply about, aren't as well equipped. My question is: How can I be assured the medium that is Hunch will protect these friends and relatives from plays on their specific vulnerabilities, rather than pushing wider the door for their exploitation?
I apologize about the long comment.
Justin Skolnick | April 17, 2009 7:44 AMGreat comments Justin, and glad you're awed -- and re: long comment, no sweat, it's an important concern to address. Hopefully I can dispel some of your worries:
In an earlier blog post I said the following, which I think bears repeating: Hunch will make money from referral fees from external sites for the subset of topics that have to do with products and services. We're not marketing things to people that they don't want, or hoarding and selling people's data. And the presence of a link has no effect on Hunch decision results.
As a longtime online community person I've got a deep sense of the importance of user trust, the privacy implications of various features, the importance of defaults for less sophisticated users -- it's been my life study since -- holy cow, The Well. The Well created a model of a civilized space which has exerted the influence for online communities since (whether they know it or not, us old folks sort of remember!)
People own their own words, their own thoughts, their own data. They control their privacy, can show what they choose to, share with whom they choose to, and participate or not. This trust was the crux of Flickr, where photos are very personal pieces of data. On Hunch, your remembered answers default to private, which is a very deliberate decision on our part, and you can delete them at any time.
I hope this helps!
Caterina | April 17, 2009 12:26 PMInterested in the API from academic standpoint.
Marc Grayson | April 17, 2009 12:46 PMAlso,
Decision trees reminds me some serious comp sci class (hope this doesn't bring back bad memories) :) I've an avid fan of artificial intelligence, natural language processing...so HUNCH dabbles in an area of "recommendations" i'd like to get my arms around.
Question. Are majority of high-level questions asked by the user...(where Hunch then asks questions to poke a little further)...just have "yes" or "no" answers?
Meaning..."should I go for a PhD"?...where final answer (hunch) is "yes" or "no".
Idea for use of Hunch API is drooling from my mouth right now.
Marc Grayson | April 17, 2009 12:58 PMCaterina, thanks for sharing this data! I cannot believe you guys already have 4.3M answered questions. That's downright amazing. Grats and keep it up!
-Paul (y! intern in your bullpen '06)
Paul Stamatiou | April 17, 2009 2:45 PMI'd be interested in the academic API.
Sean | April 17, 2009 2:54 PMDefinitely interested in the academic dataset. Could make for some interesting geographic visualizations! Cheers and keep up the good work!
Justin Davey | April 18, 2009 4:18 AMI would be interested in access to your academic API. Thanks!
Mark Watson | April 18, 2009 6:30 AMHi Caterina,
I'm an interaction designer based in NZ and conduct loads of user research. I would love access to the data you're releasing.
I have a feeling this is a product may have many unexpected uses, great work!
Caroline | April 18, 2009 2:25 PMHi Caterina,
I'm interested in the data from your API. Been fun playing around with Hunch so far.
Jon | April 20, 2009 10:25 AMI'd be interested in looking at these data.
Mark | April 20, 2009 11:29 AMI want in! Would love to see the "research" API.
David | April 20, 2009 2:35 PMI would love to have access to the academic API.
Arvind Narayanan | April 22, 2009 1:08 PMI'd be interested in the academic API.
Juho | May 5, 2009 11:02 PMHi Caterina, I would be interested in the academic API too!
Seb | June 15, 2009 11:26 PMCaterina,
I am interested in the API.
Thanks, Joseph
{ Post a comment }
I'll never forget my first zoom through flickr. You're the real thing!
Jesse | April 16, 2009 4:05 PM