Language Log

Demographic Prediction

David Palfrey has drawn my attention to Microsoft adCenter Labs' "Demographic Prediction" tool, which allows you to "use adCenter technology to predict a customer’s age, gender, and other demographic information according to his or her online behavior—that is, from search queries and webpage views."

OK, let's see if it works with basic stereotypes:

So far, so good. What about Language Log?

As I interpret what the adCenter Labs page says about this, they're predicting that the people who search for and/or read Language Log are youthful, and evenly balanced as to sex:

General Distribution is the breakdown by age of MSN Search users—based on a one-month MSN Search log—regardless of search query used.

Predicted Distribution is the predicted breakdown by age of MSN Search users for a single search query, based on the adLabs predictive model.

Well, if this is true, it's good news for the future of the field of linguistics.

But guess what? It turns out that interest in math is strongest among mature females:

This is clearly due to the role of differential equations in fending off attacks from ticked-off cavemen.

[Update -- Theo Vosse offers the results of some further explorations:

A few tries in the reliability of the adLab tool (Male - Female percentages):

http://www.google.com: 43 - 57
http://www.cnn.com: failed to predict for URL
http://www.bbc.co.uk: 61 - 39

Is Google female and the BBC male? And CNN neuter?

Simple morphological variation:

sister: 52 - 48
sisters: 28 - 72

Synonyms and plurals:
struggle: 67 - 33 struggles: 42 - 58
battle: 61 - 39 battles: 57 - 43
fight: 59 - 41 fights: 62 - 38
competition: 57 - 43 competitions: 30 - 70
match: 50 - 50 matches: 48 - 52
conflict: 50 - 50 conflicts: 30 - 70
contest: 44 - 56 contests: 32 - 68

Class variations:
cherry: 52 - 48
pear: 45 - 55
banana: 41 - 59
strawberry: 27 - 83
fruit: 35 - 65

Named entities:
travolta john: 56 - 44
travolta j: 53 - 47
j travolta: 51 - 49
travolta: 45 - 55
john travolta: 36 - 64

I wonder how you should interpret that...

So "competitions" is 30-70 female, while "struggle" is 67-33 male. Found poetry, or random noise? We report, you decide.]

Posted by Mark Liberman at September 28, 2006 07:52 AM