Radio

Sundays 8pm to 11pm on Radio 2

A Very Strombo Christmas Special

A Very Strombo Christmas Special

Throw a log on the fire, drown yourself in eggnog and turn up this year's festive special.

Listen Past Shows

 

 

Alt News
Here’s How Men And Women Talk Differently On Facebook
October 2, 2013
submit to reddit
Women
1/10 OPEN GALLERY

Heads up: the gallery contains language that may not be appropriate for all ages.

How much can you learn about someone from their postings on Facebook? It turns out, quite a lot, according to a new study from the University of Pennsylvania's Positive Psychology Center. 

The researchers examined over 15 million status updates from nearly 75,000 volunteers to explore what words and phrases were most strongly correlated with different sets of groups, like men vs. women and older vs. younger people. Each volunteer also filled out a personality questionnaire, which allowed the researchers to compare the updates of introverts vs. extroverts and more neurotic vs. more emotionally stable posters.

The results, which they turned into the word clouds in the gallery, conform almost too neatly to stereotypes: men, it turns out, were more likely to use swear words and references to specific objects ("xbox" and "ps3" were common), whereas women used more words describing their emotional state ("excited," "love you") and first-person singulars ("I"). Intriguingly, when describing romantic partners, men were more likely to use posessive determiners ("my wife," "my girlfriend") and women were more likely not to ("husband," "boyfriend").

The age-related results also showed some clear trends: the youngest group (13–18 year olds) used more emoticons and internet-speak ("idk" for "I don't know"), and the next oldest group (23–29) were more likely to talk about work ("at work," "new job"), beer and weddings, whereas the oldest group (30–65) referred frequently to children and family. Unsurprisingly, extroverts were more likely to talk about partying, and introverts about anime and the internet.

Once they fed all that data into their computational model, the researchers were able to predict a user's gender with surprising accuracy — 92 per cent — an accomplishment they consider a test case for data-driven investigations of language and personality. 

Via Popular Science

Comments

Comments on this story are moderated according to our Submission Guidelines. Comments are welcome while open. We reserve the right to close comments at any time.

Submission Policy

Note: The CBC does not necessarily endorse any of the views posted. By submitting your comments, you acknowledge that CBC has the right to reproduce, broadcast and publicize those comments or any part thereof in any manner whatsoever. Please note that comments are moderated and published according to our submission guidelines.