Modeling communities from their social mediaModeling communities from their social media Lyle Ungar...

17
Modeling communities from their social media Lyle Ungar wwbp.org University of Pennsylvania

Transcript of Modeling communities from their social mediaModeling communities from their social media Lyle Ungar...

Modeling communities from their social media

Lyle Ungar wwbp.org

University of Pennsylvania

Community-level modeling u Open vocabulary

l  Correlate word or LDA topic use with community sentiment u Model-based

l  Build language models of sentiment n  at the tweet or individual level

l  Apply them to new tweets or people n  Group by community

l  (Sometimes) correlate with outcome

2

Measuring personality in tweets u Create regression models for personality u Use these to estimate scores on tweets

collected on the county level

u Agreeableness u Conscientiousness u Extroversion u Neuroticism u Openness to experience

Agreeableness

Conscientiousness

Extraversion

Neuroticism

Openness to experience

9

Satisfaction with Life Prediction

10

Controls • median age • sex (percentage female) • minorities (percentage black and Hispanic). • median household income (log-transformed) • educational attainment (% high school grads or higher; % BS or higher).

county-level predictions

Satisfaction with Life

11 survey predicted

Life Satisfaction Words

12

13

14

Twitter predicts cardiovascular disease

Twitter predicts cardiovascular disease

15

Cardiovascular Disease Words

16

Person & Community Conclusions u Language reveals demographics, personality

l  Also psychopathy, emotion, SES, political orientation, happiness, depression, health, disease

u Language models generalize l  Across individuals l  From Facebook to Twitter l  From Individuals to Counties

u Language allows data-driven hypothesis generation l  As well as hypothesis testing

17