SNAP: Social Media and Network Analy5cs for Public Health

•  HenryKautz•  GoergenIns1tuteforDataScience,UniversityofRochester

•  Idea•  Socialmediausers=distributedsensornetwork

•  Challenge•  Discoveringsignalinnoise

• Method•  Machinelearning

•  Applica1ons•  Trackinginfluenza•  Measuringalcoholuse•  Improvingfoodsafety

!me-lapseheatmapoftweetsfromNYC

Analyzing Tweets • Goal:findraretweetsaboutspecificdiseasesymptoms(1/50,000)

•  Previousapproach:keywords•  Problems:“sickofhomework”,“undertheweather”

• Ourapproach:machinelearning•  Use“MechanicalTurk”workerstocreatetrainingdata•  98%accuracy

SickTweetsMachine

LearningSystem

TrainingData

Contains“sneeze”?“sick”?“1red”?

Twi@erhealth Usinga“flusymptom”classifier,wecan:• Measureflulevelsaccuratelyandquickly• Predictriskofpar1cularuserscatchingtheflu• Discovercorrela1onsbetweenfluandfactorssuchasairpollu1on

nEmesis

•  Listentotweetstofindpossiblecasesoffoodpoisoning• Useresultstopriori1zerestaurantinspec1ons•  3monthtrialisLasVegas:provedeffec1veindoubleblindtrial• CDCfunding5yearexpansion

GeoDrink

• UnderstandingpaZernsofalcoholuseincommuni1es•  Inferloca1onsofusers’homesandtheexact1meandplaceofdrinking