Twitter and Alcohol - BrightonSEO Pressentation
-
Upload
daniel-kershaw -
Category
Social Media
-
view
340 -
download
0
description
Transcript of Twitter and Alcohol - BrightonSEO Pressentation
I'm drunk ... Karaoke Drunk
#DONTTRYSPELLINGKSRAOKDR
UN
Highwire CDT
Lancaster University
SCC
Lancaster University
Managment Science
Lancaster University
Monitoring Regional Alcohol ConsumptionThrough Social Media
Daniel Kershaw
Matthew Rowe
Patrick Stacey
People Like toDrink
UK Alcohol Consumption fromthe 1900'S
Varying Rates of Harm
Current Data CollectionMethonds
Quantity Frequency Questionnaires (QF)Time Line Method (TL)Time consumingExpensiveData is only a snapshot of the past
Data Collection ErrorsSelective reportingRecall biasAccidental under-estimation by up to 40%
People Like to Tweet
Why Do People Use TwitterMinimal EffortMobile and pervasivePeople-based RSS feedsBroadcast Nature of TwitterKeeping in touch with friends and familyGathering information / Seeking help / Releasing emotionalstress
Previous WorkMonitoring Flu Spreading Though Twitter - Culotta, A. (2010)Social Media to Track Depression on a Global Scale - DeChoudhury, M., Counts, S., & Horvitz, E. (2013)Stock Market Prediction Through Sentiment Analysis - Bollen,Mao, Zeng. (2011)Detecting Earthquakes Through Peoples Tweets - Sakaki, T.,Okazaki, M., & MATSUO, Y. (2010)
Twitter as aSpatio-temporalSense Network
ResearchQuestion
Is it possible to characterise and model UKalcohol consumption patterns of alcohol on
social media data such as Twitter, and if so isthere a variation across geographical location in
drinking patterns and terminology usage?
Grounded TruthHealth and Social Care Infomation Center (HSCIC)Statistics on Alcohol ReportLooking for Daily Granularity
Sunday Monday Tuesday Wednesday Thursday Friday Saturday
5
10
15
20
25
30
35
Day of the Week
% o
f res
pond
ent
Combined16 - 2425 - 4445 - 6465 +
plotly - data and graph »
TwitterStreaming API
Bounding Box
31.6 million Tweets over 6 week period700,000 tweets/daily500 tweets/minute8 tweets/second40Gb of Data to process
MethodSimple Average Keyword Signal Analysis
KeywordsDrunk Wine BeerHangover Hungover WastedPissed
@JeremyClarkson are you dead? pic.twitter.com/BsT7SlYAPvAdzy @iliffe25
@iliffe25 A bit pissed. But not dead11:52 PM - 10 Jun 2014
Jeremy Clarkson @JeremyClarkson
Follow
132 RETWEETS 210 FAVORITES
10 Jun
My cat is sad because his mate got drunk at a strip club last night & is now vomiting in a quiet corner of the house. 7:30 AM - 12 Jun 2014
WHY MY CAT IS SAD @MYSADCAT
Follow
231 RETWEETS 361 FAVORITES
Write drunk. Edit sober.— Shit Academics Say
(@AcademicsSay) June 18, 2014
The Math(s)SMAI(T, M) = s(t,M)*t!T
|T|
s(t, M) = c(t,m)*m!Mtokens(t)<< <<
c(t, m) = f (w, m)*w!tokens(t)
f : W × M → {0, 1}
Groupings
31.6 million tweets becomes 252.8 million data points - 320 Gbto process
National → North West → LA → LA1
Lets Look at theData
I'll Drink to That
Week 1 Week 2 Week 3 Week 4 Week 5 Week 6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
1
Week of Study
Corr
olat
ion
National UKNorth WestYorkshire & HumbersideGreater LondonSouth WestSouth EastNorthern IrelandWest MidlandsChannel IslandsHome CountiesScotland (North)East EnglandScotland (South & Central)Wales (South)Wales (North)East MidlandsNorth EastAvrage
plotly - data and graph »
52 54 56 58 60 62 64 66
500μ
550μ
600μ
650μ
700μ
Drank last week (% of poppulation)
Avra
ge S
MAI
plotly - data and graph »
The ParablesNot to replace current methods, only too supplement themWord Sence DisambiguationTwitterology
Future WorkOpen vocabulary methodLooking at conversations around alcoholSmoothing of results using demographic modeling
To take homeThe ability to track underlying social trendsWe can detect the trend with high correlation to nationalstatisticsSimple to implement