Tweets Automan Predict Automan CSE EconomicsISR Social Media

Tweets Automan Predict Automan CSE Economics/ISR

Social Media

7. 6 Billion Tweets

7. 6 Billion Tweets June 2011 - July 2012

7. 6 Billion Tweets June 2011 - July 2012

Automan

Prediction

US Initial Claims for Unemployment Insurance Chart

1 Week

Tweets 3 Categories of Signals (16 Signals) 1 Week Job Search Lost job Unemployment “find a job” “just fired” “unemployment” … … …

Tweets 3 Categories of Signals (16 Signals) 1 Week Job Search Lost job Unemployment “find a job” “just fired” “unemployment” … … … Auto-regression + Seasonal factor Predictive Signal

Initial Claims (SA) versus Twitter Signal 430 K Ground truth Initial Claims 410 K 390 K 370 K Automan predictive signal 350 K 2011 2012 Time

Release

Release Average Economists

Release Predict Average Economists

Release Predict Surprise! Average Economists

Initial Claims 20 K Surprise 0 Automan prediction -20 K 2011 2012 Time

Credit Dolan Antenucci, Michael Cafarella, Erdong Li, Shaobo Liu Computer Science Victor Bittorf, Arun Kumar, Christopher Ré, Ce Zhang Computer Science Sponsors Special Thanks Margaret Levenstein, Matthew Shapiro ISR and Economics Pete Cafarella Qiaozhu Mei Vijairam Parasuraman

Automan A prototype tool for performing economic predictions using social media data.

Automan A prototype tool for performing economic predictions using social media data. Result Automan can generate a signal from social media data that tracks the true number of initial unemployment benefits claims.

Automan A prototype tool for performing economic predictions using social media data. Result 1 Automan can generate a signal from social media data that tracks the true number of initial unemployment benefits claims. Result 2 Automan can explain roughly 1/3 of the error in economists’ predictions.