Members: Raghuram Krishnamachari Manish Maheshwari Maryam El Kherba Guided by: Prof. Alan Mislove
Flu Prediction / Activity CDC Flu Activity Reports Influenza like Illness (ILI) for each region Google Flu Trends Aggregates search data to estimate flu activity Our experiment (Twitter) Analyze Twitter data (tweets) to estimate flu activity
Google Flu Trends CDC’s ILI data VS Google Flu Trends
Google Flu Trends Vs Twitter 12000 HHS Region 1 (CT, ME, MA, NH, RI, VT) HHS Region 2 (NJ, NY) 10000 HHS Region 3 (DE, DC, MD, PA, 8000 VA, WV) HHS Region 4 (AL, FL, GA, KY, MS, NC, SC, TN) HHS Region 5 (IL, IN, MI, MN, 6000 OH, WI) HHS Region 6 (AR, LA, NM, OK, 4000 TX) HHS Region 7 (IA, KS, MO, NE) 2000 HHS Region 8 (CO, MT, ND, SD, UT, WY) HHS Region 9 (AZ, CA, HI, NV) 0 HHS Region 10 (AK, ID, OR, WA) United States 0.009 Region 1 0.008 0.007 Region 2 0.006 Region 3 0.005 Region 4 0.004 Region 5 0.003 Region 6 0.002 0.001 Region 7 0 Region 8 Region 9 Region 10
Google Flu Trends Vs Twitter 7000 6000 5000 4000 3000 G-R3 2000 T-R3 1000 0 8000 7000 6000 5000 4000 3000 G-R9 2000 T-R9 1000 0
Tweets, Phrases "having a cold" 4 "have a cold“ 7 "feel feverish" "flu" 5 "headache" "flu" 8 "sick" "flu" 9 "flu" "fever“ 5 "came down with the flu" 7 "chills" "flu" 7 "catching the flu" 6 "cough" "flu" 6 "fatigue" "flu" 8 "weakness" "flu" 6 "flu like symptoms" 4 "runny nose" "flu" 5 "sore throat" "flu" 7 "stomach ache" "flu" 6 "stuffy nose" "flu" 6 "tiredness" "flu" 4 "vomiting" "flu" 4 "watery eyes" "flu" 6 "body hurts" "flu" 7
Process • Filter flu tweets from twitter data Filter • Store data for each state (FIPS) • Count flu tweets (weekly) Count • Count total tweets (weekly) • Ratio of flu related to total tweets Plot • Compare against Google/CDC
Implementation Linux bash shell script Filtering find fips -name "*.gz" -exec zcat {} \; | grep "$1" Counting find … -exec zcat {} \; | awk ‘{ print $3 }' | awk '{ print $3 " " $2 " " $6 } sort -k 3n -k 2M -k 1n | uniq -c Plotting pr -mft -s, dates.txt NJ.tot NY.tot > RE2.tot Microsoft Excel
Challenges Filtering Phrases that express flu symptoms Processing time Segregation based on location Counting Processing time Storage format Plotting Lack of consistent CDC data Handling of large numeric data
Future Better prediction algorithm Live Tweet monitoring Flu propagation Facebook application
Recommend
More recommend