what to expect when the unexpected happens: social media communications across crises
TRANSCRIPT
What to Expect When The Unexpected Happens:
Social Media Communications Across Crises
!Alexandra Olteanu Sarah Vieweg, Carlos Castillo
Leetaru et al. "Mapping the global Twitter heartbeat: The geography of Twitter." First Monday (2013)
What are the similarities and differences in Twitter communications that take place during different crisis events, according to specific characteristics of such
events?
What are the similarities and differences in Twitter communications that take place during different crisis events, according to specific characteristics of such
events?
What are the similarities and differences in Twitter communications that take place during different crisis events, according to specific characteristics of such
events?
Analysis Framework
Crisis Dimensions
Temporal Development
Geograp
hic Sprea
d
Cris
is T
ype
natural
human-induced
Analysis Framework
Crisis Dimensions
Temporal Development
Geograp
hic Sprea
d
Cris
is T
ype
natural
human-induced
Analysis Framework
Crisis Dimensions
Content Dimensions
Eyewitness Government NGOs Media Business Outsiders
Affected individuals
Infrastructure Damage
DonationsCaution &
AdviceSympathy
Other
SourceType
Analysis Framework
Crisis Dimensions
Content Dimensions
not informative informative
Eyewitness Government NGOs Media Business Outsiders
Affected individuals
Infrastructure Damage
DonationsCaution &
AdviceSympathy
Other
SourceType
Analysis Framework
Crisis Dimensions
Content Dimensions
Data Collection
Data Annotation
Data Analysis
Costa Rica earthquake’12
Manila floods’13
Singapore haze’13
Queensland floods’13
Typhoon Pablo’12
Australia bushfire’13
Italy earthquakes’12
Sardinia floods’13
Philipinnes floods’12
Alberta floods’13
Typhoon Yolanda’13
Colorado floods’13
Guatemala earthquake’12
Colorado wildfires’12
Bohol earthquake’13
NY train crash’13
Boston bombings’13
LA airport shootings’13
West Texas explosion’13
Russia meteor’13
Savar building collapse’13
Lac Megantic train crash’13
Venezuela refinery’12
Glasgow helicopter crash’13
Spain train crash’13
Brazil nightclub fire’13
0
10
20
30
40
50
60
70
80
90
100Caution & AdviceAffected Ind.Infrast. & UtilitiesDonat. & Volun.SympathyOther Useful Info.
Med
ia
Out
side
rs
Eye
witn
ess
Gov
ernm
ent
NG
Os
Bus
ines
s
Other Useful Info.
Sympathy
Affected Ind.
Donat. & Volun.
Caution & Advice
Infrast. & Utilities
0
5%
10%
>15%
Savar building collapse’13LA airport shootings’13
NY train crash’13Russia meteor’13
Colorado wildfires’12Guatemala earthquake’12
Glasgow helicopter crash’13West Texas explosion’13
Lac Megantic train crash’13Venezuela refinery’12Bohol earthquake’13Boston bombings’13
Brazil nightclub fire’13Spain train crash’13
Manila floods’13Alberta floods’13
Philipinnes floods’12Typhoon Yolanda’13
Costa Rica earthquake’12Singapore haze’13
Italy earthquakes’12Australia bushfire’13
Queensland floods’13Colorado floods’13Sardinia floods’13Typhoon Pablo’12
Analysis Framework
Crisis Dimensions
Content Dimensions
Data Collection
Data Annotation
Data Analysis
Step 1 Step 2 Step 3 Step 4 Step 5
Step 1: Events TypologyNatural Human-Induced
Progressive
Focalized Epidemics Demonstrations, Riots
Diffused Floods, Storms, Wildfires Wars
Instantaneous
Focalized Meteorites, Landslides Train accidents, Building collapse
Diffused Earthquakes Large-scale industrial accidents
Step 2: Content Dimensions ☞ Informativeness
Informative: useful information, situational information, etc.
Not informative: prayers, trolling, spam, humor, rumor, off-topic, etc.
☞ Source of information Primary sources
Eyewitness: citizen reporters, local individuals, direct experience
Secondary & tertiary sources Government: authorities, police & fire services, public institutions NGOs: non-profit org., non-governmental org., faith-based org. Business: comercial org., enterprises, for-profit corporation Media: news org., journalists, news media!Outsiders: remote crowd, non-locals, sympathizers
☞ Type of information !
!
!
!
!
!
!
!
!
Step 2: Content Dimensions ☞ Informativeness
Informative: useful information, situational information, etc.
Not informative: prayers, trolling, spam, humor, rumor, off-topic, etc.
☞ Source of information Primary sources
Eyewitness: citizen reporters, local individuals, direct experience
Secondary & tertiary sources Government: authorities, police & fire services, public institutions NGOs: non-profit org., non-governmental org., faith-based org. Business: comercial org., enterprises, for-profit corporation Media: news org., journalists, news media!Outsiders: remote crowd, non-locals, sympathizers
☞ Type of information !
!
!
!
!
!
!
Step 2: Content Dimensions ☞ Informativeness
Informative: useful information, situational information, etc.
Not informative: prayers, trolling, spam, humor, rumor, off-topic, etc.
☞ Source of information Primary sources
Eyewitness: citizen reporters, local individuals, direct experience
Secondary & tertiary sources Government: authorities, police & fire services, public institutions NGOs: non-profit org., non-governmental org., faith-based org. Business: comercial org., enterprises, for-profit corporation Media: news org., journalists, news media!Outsiders: remote crowd, non-locals, sympathizers
☞ Type of information !
!
!
!
!
!
!
!
!
Step 2: Content Dimensions ☞ Informativeness
Informative: useful information, situational information, etc.
Not informative: prayers, trolling, spam, humor, rumor, off-topic, etc.
☞ Source of information Primary sources
Eyewitness: citizen reporters, local individuals, direct experience
Secondary & tertiary sources Government: authorities, police & fire services, public institutions NGOs: non-profit org., non-governmental org., faith-based org. Business: comercial org., enterprises, for-profit corporation Media: news org., journalists, news media!Outsiders: remote crowd, non-locals, sympathizers
☞ Type of information !
Step 2: Content Dimensions ☞ Informativeness
Informative: useful information, situational information, etc.
Not informative: prayers, trolling, spam, humor, rumor, off-topic, etc.
☞ Source of information Primary sources
Eyewitness: citizen reporters, local individuals, direct experience
Secondary & tertiary sources Government: authorities, police & fire services, public institutions NGOs: non-profit org., non-governmental org., faith-based org. Business: comercial org., enterprises, for-profit corporation Media: news org., journalists, news media!Outsiders: remote crowd, non-locals, sympathizers
☞ Type of information !
!
!
!
!
!
!
!
!
Step 2: Content Dimensions ☞ Informativeness
Informative: useful information, situational information, etc.
Not informative: prayers, trolling, spam, humor, rumor, off-topic, etc.
☞ Source of information Primary sources
Eyewitness: citizen reporters, local individuals, direct experience
Secondary & tertiary sources Government: authorities, police & fire services, public institutions NGOs: non-profit org., non-governmental org., faith-based org. Business: comercial org., enterprises, for-profit corporation Media: news org., journalists, news media!Outsiders: remote crowd, non-locals, sympathizers
☞ Type of information !
!
!
!
!
!
!
!
!
Affected individuals!casualties; people missing, found, trapped, seen; reports about self
Infrastructure & utilities!road closures, collapsed structure, water sanitation, services
Donations & volunteering requesting help, proposing relief, relief coordination, shelter needed
Caution & advice!predicting or forecasting, instructions to handle certain situations
Sympathy & emotional support!thanks, gratitude, prayers, condolences, emotion-related
Other useful information!meta-discussions, flood level, wind, visibility, weather
Step 3: Data Collection ☞ Twitter base-sample* ☞ 2012 & 2013 ☞ ~1% random sample of Twitter public stream ☞ ~130+ million tweets per month
☞Keyword-based searches ☞ proper names of affected location ☞ manila floods, boston bombings, #newyork derailment
☞ proper names of meteorological phenomena ☞ sandy hurricane, typhoon yolanda
☞ promoted hashtags ☞ #SafeNow, #RescuePH, #ReliefPH
☞ 26 crisis events ☞ 14 countries and 8 languages ☞ 12 different hazard types ☞ earthquake, wildfire, floods, bombings, shootings, etc. !
☞ 15 instantaneous crises ☞ 15 diffused crises
* https://archive.org/details/twitterstream
Step 3: Data Collection ☞ Twitter base-sample* ☞ 2012 & 2013 ☞ ~1% random sample of Twitter public stream ☞ ~130+ million tweets per month
☞Keyword-based searches ☞ proper names of affected location ☞ manila floods, boston bombings, #newyork derailment
☞ proper names of meteorological phenomena ☞ sandy hurricane, typhoon yolanda
☞ promoted hashtags ☞ #SafeNow, #RescuePH, #ReliefPH
☞ 26 crisis events ☞ 14 countries and 8 languages ☞ 12 different hazard types ☞ earthquake, wildfire, floods, bombings, shootings, etc. !
☞ 15 instantaneous crises ☞ 15 diffused crises
* https://archive.org/details/twitterstream
Step 3: Data Collection ☞ Twitter base-sample* ☞ 2012 & 2013 ☞ ~1% random sample of Twitter public stream ☞ ~130+ million tweets per month
☞Keyword-based searches ☞ proper names of affected location ☞ manila floods, boston bombings, #newyork derailment
☞ proper names of meteorological phenomena ☞ sandy hurricane, typhoon yolanda
☞ promoted hashtags ☞ #SafeNow, #RescuePH, #ReliefPH
☞ 26 crisis events ☞ 14 countries and 8 languages ☞ 12 different hazard types ☞ earthquake, wildfire, floods, bombings, shootings, etc. !
☞ 15 instantaneous crises ☞ 15 diffused crises
* https://archive.org/details/twitterstream
Step 3: Data Collection ☞ Twitter base-sample* ☞ 2012 & 2013 ☞ ~1% random sample of Twitter public stream ☞ ~130+ million tweets per month
☞Keyword-based searches ☞ proper names of affected location ☞ manila floods, boston bombings, #newyork derailment
☞ proper names of meteorological phenomena ☞ sandy hurricane, typhoon yolanda
☞ promoted hashtags ☞ #SafeNow, #RescuePH, #ReliefPH
☞ 26 crisis events ☞ 14 countries and 8 languages ☞ 12 different hazard types ☞ earthquake, wildfire, floods, bombings, shootings, etc. !
☞ 15 instantaneous crises ☞ 15 diffused crises
* https://archive.org/details/twitterstream
Step 3: Data Collection ☞ Twitter base-sample* ☞ 2012 & 2013 ☞ ~1% random sample of Twitter public stream ☞ ~130+ million tweets per month
☞Keyword-based searches ☞ proper names of affected location ☞ manila floods, boston bombings, #newyork derailment
☞ proper names of meteorological phenomena ☞ sandy hurricane, typhoon yolanda
☞ promoted hashtags ☞ #SafeNow, #RescuePH, #ReliefPH
☞ 26 crisis events ☞ 14 countries and 8 languages ☞ 12 different hazard types ☞ earthquake, wildfire, floods, bombings, shootings, etc. !
☞ 15 instantaneous crises ☞ 15 diffused crises
* https://archive.org/details/twitterstream
Step 3: Data Collection ☞ Twitter base-sample* ☞ 2012 & 2013 ☞ ~1% random sample of Twitter public stream ☞ ~130+ million tweets per month
☞Keyword-based searches ☞ proper names of affected location ☞ manila floods, boston bombings, #newyork derailment
☞ proper names of meteorological phenomena ☞ sandy hurricane, typhoon yolanda
☞ promoted hashtags ☞ #SafeNow, #RescuePH, #ReliefPH
☞ 26 crisis events ☞ 14 countries and 8 languages ☞ 12 different hazard types ☞ earthquake, wildfire, floods, bombings, shootings, etc. !
☞ 15 instantaneous crises ☞ 15 diffused crises
* https://archive.org/details/twitterstream
Step 4: Data Annotation ☞ ~1000 tweets per crisis ☞ Informativeness ☞ Source of information ☞ Type of information
!☞ Crowdsource workers from the affected countries
Step 5: Data Analysis ☞ Content types/sources vs. crisis dimensions
☞ Interplay between types and sources
☞ Crisis similarity
☞ Temporal aspects
What: Types of InformationInfrastructure and utilities: 7% on average (min. 0%, max. 22%)!
most prevalent in diffused crises, in particular during floods!Caution and advice: 10% on average (min. 0%, max. 34%)!
least prevalent in instantaneous & human-induced disasters!Donations and volunteering: 10% on average (min. 0%, max. 44%)!
most prevalent in natural disasters
Costa Rica earthquake’12
Manila floods’13
Singapore haze’13
Queensland floods’13
Typhoon Pablo’12
Australia bushfire’13
Italy earthquakes’12
Sardinia floods’13
Philipinnes floods’12
Alberta floods’13
Typhoon Yolanda’13
Colorado floods’13
Guatemala earthquake’12
Colorado wildfires’12
Bohol earthquake’13
NY train crash’13
Boston bombings’13
LA airport shootings’13
West Texas explosion’13
Russia meteor’13
Savar building collapse’13
Lac Megantic train crash’13
Venezuela refinery’12
Glasgow helicopter crash’13
Spain train crash’13
Brazil nightclub fire’13
0
10
20
30
40
50
60
70
80
90
100Caution & AdviceAffected Ind.Infrast. & UtilitiesDonat. & Volun.SympathyOther Useful Info.
Affected individuals: 20% on average (min. 5%, max. 57%)!most prevalent in human-induced, focalized & instantaneous crises!
Sympathy and emotional support: 20% on average (min. 3%, max. 52%)!most prevalent in instantaneous crises!
Other useful information: 32% on average (min. 7%, max. 59%)!least prevalent in diffused crises
Costa Rica earthquake’12
Manila floods’13
Singapore haze’13
Queensland floods’13
Typhoon Pablo’12
Australia bushfire’13
Italy earthquakes’12
Sardinia floods’13
Philipinnes floods’12
Alberta floods’13
Typhoon Yolanda’13
Colorado floods’13
Guatemala earthquake’12
Colorado wildfires’12
Bohol earthquake’13
NY train crash’13
Boston bombings’13
LA airport shootings’13
West Texas explosion’13
Russia meteor’13
Savar building collapse’13
Lac Megantic train crash’13
Venezuela refinery’12
Glasgow helicopter crash’13
Spain train crash’13
Brazil nightclub fire’13
0
10
20
30
40
50
60
70
80
90
100Caution & AdviceAffected Ind.Infrast. & UtilitiesDonat. & Volun.SympathyOther Useful Info.
What: Types of Information
Who: Sources of Information
Singapore haze’13
Philipinnes floods’12
Alberta floods’13
Manila floods’13
Queensland floods’13
Typhoon Pablo’12
Italy earthquakes’12
Australia bushfire’13
Colorado floods’13
Colorado wildfires’12
Bohol earthquake’13
Costa Rica earthquake’12
LA airport shootings’13
Venezuela refinery’12
West Texas explosion’13
Sardinia floods’13
Spain train crash’13
Guatemala earthquake’12
Brazil nightclub fire’13
Boston bombings’13
Glasgow helicopter crash’13
Russia meteor’13
Lac Megantic train crash’13
Typhoon Yolanda’13
Savar building collapse’13
NY train crash’13
0
10
20
30
40
50
60
70
80
90
100EyewitnessGovernmentNGOsBusinessMediaOutsiders
Business: 2% on average (min. 0%, max. 9%)!most prevalent in diffused crises!
NGOs: 4% on average (min. 0%, max. 17%)!most prevalent in natural disasters, in particular during typhoons & floods!
Government: 5% on average (min. 1%, max. 13%)!most prevalent in natural, progressive & diffused crises
Eyewitness accounts: 9% on average (min. 0%, max. 54%)!most prevalent in progressive & diffused crises!
Outsiders: 38% on average (min. 3%, max. 65%) !least in the Singapore Haze crisis!
Traditional & Internet media: 42% on average (min. 18%, max. 77%) !most prevalent in instantaneous crises, which make the “breaking news”
Who: Sources of Information
Singapore haze’13
Philipinnes floods’12
Alberta floods’13
Manila floods’13
Queensland floods’13
Typhoon Pablo’12
Italy earthquakes’12
Australia bushfire’13
Colorado floods’13
Colorado wildfires’12
Bohol earthquake’13
Costa Rica earthquake’12
LA airport shootings’13
Venezuela refinery’12
West Texas explosion’13
Sardinia floods’13
Spain train crash’13
Guatemala earthquake’12
Brazil nightclub fire’13
Boston bombings’13
Glasgow helicopter crash’13
Russia meteor’13
Lac Megantic train crash’13
Typhoon Yolanda’13
Savar building collapse’13
NY train crash’13
0
10
20
30
40
50
60
70
80
90
100EyewitnessGovernmentNGOsBusinessMediaOutsiders
Events Similarity: Information Types
Savar building collapse’13LA airport shootings’13
NY train crash’13Russia meteor’13
Colorado wildfires’12Guatemala earthquake’12
Glasgow helicopter crash’13West Texas explosion’13
Lac Megantic train crash’13Venezuela refinery’12Bohol earthquake’13Boston bombings’13
Brazil nightclub fire’13Spain train crash’13
Manila floods’13Alberta floods’13
Philipinnes floods’12Typhoon Yolanda’13
Costa Rica earthquake’12Singapore haze’13
Italy earthquakes’12Australia bushfire’13
Queensland floods’13Colorado floods’13Sardinia floods’13Typhoon Pablo’12
lower similarity
Events Similarity: Information Types
Savar building collapse’13LA airport shootings’13
NY train crash’13Russia meteor’13
Colorado wildfires’12Guatemala earthquake’12
Glasgow helicopter crash’13West Texas explosion’13
Lac Megantic train crash’13Venezuela refinery’12Bohol earthquake’13Boston bombings’13
Brazil nightclub fire’13Spain train crash’13
Manila floods’13Alberta floods’13
Philipinnes floods’12Typhoon Yolanda’13
Costa Rica earthquake’12Singapore haze’13
Italy earthquakes’12Australia bushfire’13
Queensland floods’13Colorado floods’13Sardinia floods’13Typhoon Pablo’12
lower similarity
Events Similarity: Information Types
Savar building collapse’13LA airport shootings’13
NY train crash’13Russia meteor’13
Colorado wildfires’12Guatemala earthquake’12
Glasgow helicopter crash’13West Texas explosion’13
Lac Megantic train crash’13Venezuela refinery’12Bohol earthquake’13Boston bombings’13
Brazil nightclub fire’13Spain train crash’13
Manila floods’13Alberta floods’13
Philipinnes floods’12Typhoon Yolanda’13
Costa Rica earthquake’12Singapore haze’13
Italy earthquakes’12Australia bushfire’13
Queensland floods’13Colorado floods’13Sardinia floods’13Typhoon Pablo’12 Dominated by
natural, diffused and progressive
lower similarity
Events Similarity: Information Types
Savar building collapse’13LA airport shootings’13
NY train crash’13Russia meteor’13
Colorado wildfires’12Guatemala earthquake’12
Glasgow helicopter crash’13West Texas explosion’13
Lac Megantic train crash’13Venezuela refinery’12Bohol earthquake’13Boston bombings’13
Brazil nightclub fire’13Spain train crash’13
Manila floods’13Alberta floods’13
Philipinnes floods’12Typhoon Yolanda’13
Costa Rica earthquake’12Singapore haze’13
Italy earthquakes’12Australia bushfire’13
Queensland floods’13Colorado floods’13Sardinia floods’13Typhoon Pablo’12 Dominated by
natural, diffused and progressive
Dominated by human-induced,
focalized and instantaneous
lower similarity
Singapore haze’13Philipinnes floods’12
Alberta floods’13Manila floods’13
Italy earthquakes’12Australia bushfire’13Colorado wildfires’12
Colorado floods’13Queensland floods’13
Typhoon Pablo’12Typhoon Yolanda’13
Brazil nightclub fire’13Russia meteor’13
Boston bombings’13Glasgow helicopter crash’13
Bohol earthquake’13Venezuela refinery’12
Sardinia floods’13West Texas explosion’13
NY train crash’13Costa Rica earthquake’12
LA airport shootings’13Savar building collapse’13
Lac Megantic train crash’13Guatemala earthquake’12
Spain train crash’13
lower similarity
Events Similarity: Information Sources
Singapore haze’13Philipinnes floods’12
Alberta floods’13Manila floods’13
Italy earthquakes’12Australia bushfire’13Colorado wildfires’12
Colorado floods’13Queensland floods’13
Typhoon Pablo’12Typhoon Yolanda’13
Brazil nightclub fire’13Russia meteor’13
Boston bombings’13Glasgow helicopter crash’13
Bohol earthquake’13Venezuela refinery’12
Sardinia floods’13West Texas explosion’13
NY train crash’13Costa Rica earthquake’12
LA airport shootings’13Savar building collapse’13
Lac Megantic train crash’13Guatemala earthquake’12
Spain train crash’13
lower similarity
Events Similarity: Information Sources
Singapore haze’13Philipinnes floods’12
Alberta floods’13Manila floods’13
Italy earthquakes’12Australia bushfire’13Colorado wildfires’12
Colorado floods’13Queensland floods’13
Typhoon Pablo’12Typhoon Yolanda’13
Brazil nightclub fire’13Russia meteor’13
Boston bombings’13Glasgow helicopter crash’13
Bohol earthquake’13Venezuela refinery’12
Sardinia floods’13West Texas explosion’13
NY train crash’13Costa Rica earthquake’12
LA airport shootings’13Savar building collapse’13
Lac Megantic train crash’13Guatemala earthquake’12
Spain train crash’13
lower similarity
Dominated by instantaneous, focalized and human-induced
Events Similarity: Information Sources
Singapore haze’13Philipinnes floods’12
Alberta floods’13Manila floods’13
Italy earthquakes’12Australia bushfire’13Colorado wildfires’12
Colorado floods’13Queensland floods’13
Typhoon Pablo’12Typhoon Yolanda’13
Brazil nightclub fire’13Russia meteor’13
Boston bombings’13Glasgow helicopter crash’13
Bohol earthquake’13Venezuela refinery’12
Sardinia floods’13West Texas explosion’13
NY train crash’13Costa Rica earthquake’12
LA airport shootings’13Savar building collapse’13
Lac Megantic train crash’13Guatemala earthquake’12
Spain train crash’13
lower similarity
Dominated by instantaneous, focalized and human-induced
Events Similarity: Information Sources
Singapore haze’13Philipinnes floods’12
Alberta floods’13Manila floods’13
Italy earthquakes’12Australia bushfire’13Colorado wildfires’12
Colorado floods’13Queensland floods’13
Typhoon Pablo’12Typhoon Yolanda’13
Brazil nightclub fire’13Russia meteor’13
Boston bombings’13Glasgow helicopter crash’13
Bohol earthquake’13Venezuela refinery’12
Sardinia floods’13West Texas explosion’13
NY train crash’13Costa Rica earthquake’12
LA airport shootings’13Savar building collapse’13
Lac Megantic train crash’13Guatemala earthquake’12
Spain train crash’13
lower similarity
Dominated by instantaneous, focalized and human-induced
Events Similarity: Information Sources
Singapore haze’13Philipinnes floods’12
Alberta floods’13Manila floods’13
Italy earthquakes’12Australia bushfire’13Colorado wildfires’12
Colorado floods’13Queensland floods’13
Typhoon Pablo’12Typhoon Yolanda’13
Brazil nightclub fire’13Russia meteor’13
Boston bombings’13Glasgow helicopter crash’13
Bohol earthquake’13Venezuela refinery’12
Sardinia floods’13West Texas explosion’13
NY train crash’13Costa Rica earthquake’12
LA airport shootings’13Savar building collapse’13
Lac Megantic train crash’13Guatemala earthquake’12
Spain train crash’13
lower similarity
Dominated by instantaneous, focalized and human-induced
Dominated by natural, diffused and progressive
Events Similarity: Information Sources
Who Says What?
Med
ia
Out
side
rs
Eye
witn
ess
Gov
ernm
ent
NG
Os
Bus
ines
s
Other Useful Info.
Sympathy
Affected Ind.
Donat. & Volun.
Caution & Advice
Infrast. & Utilities
0
5%
10%
>15%
Who Says What?
Med
ia
Out
side
rs
Eye
witn
ess
Gov
ernm
ent
NG
Os
Bus
ines
s
Other Useful Info.
Sympathy
Affected Ind.
Donat. & Volun.
Caution & Advice
Infrast. & Utilities
0
5%
10%
>15%
Who Says What?
Med
ia
Out
side
rs
Eye
witn
ess
Gov
ernm
ent
NG
Os
Bus
ines
s
Other Useful Info.
Sympathy
Affected Ind.
Donat. & Volun.
Caution & Advice
Infrast. & Utilities
0
5%
10%
>15%
Who Says What?
Med
ia
Out
side
rs
Eye
witn
ess
Gov
ernm
ent
NG
Os
Bus
ines
s
Other Useful Info.
Sympathy
Affected Ind.
Donat. & Volun.
Caution & Advice
Infrast. & Utilities
0
5%
10%
>15%
Who Says What?
Med
ia
Out
side
rs
Eye
witn
ess
Gov
ernm
ent
NG
Os
Bus
ines
s
Other Useful Info.
Sympathy
Affected Ind.
Donat. & Volun.
Caution & Advice
Infrast. & Utilities
0
5%
10%
>15%
Temporal Distribution: Types
12h 24h 36h 48h … several days
peak
Caution & Advice
Sympathy & Support
Temporal Distribution: Types
12h 24h 36h 48h … several days
peak
Caution & Advice
Sympathy & Support
Infrastructure & Utilities
Temporal Distribution: Types
12h 24h 36h 48h … several days
peak
Caution & Advice
Sympathy & Support
Affected Individuals
Infrastructure & Utilities
Temporal Distribution: Types
12h 24h 36h 48h … several days
peak
Caution & Advice
Sympathy & Support
Affected Individuals
Infrastructure & Utilities
Other Specific Info.
Temporal Distribution: Types
12h 24h 36h 48h … several days
peak
Caution & Advice
Sympathy & Support
Affected Individuals
Infrastructure & Utilities
Other Specific Info.
Donations & Volunteering
Temporal Distribution: Sources
12h 24h 36h 48h … several days
peak
Eyewitness
Government
Progressive
Temporal Distribution: Sources
12h 24h 36h 48h … several days
peak
Eyewitness
Government Media
Progressive
Temporal Distribution: Sources
12h 24h 36h 48h … several days
peak
Eyewitness
Government
Outsiders
Media
Progressive
Temporal Distribution: Sources
12h 24h 36h 48h … several days
peak
Eyewitness
Government
NGOsOutsiders
Media
Progressive
Temporal Distribution: Sources
12h 24h 36h 48h … several days
peak
Eyewitness
Government
Business
NGOsOutsiders
Media
Progressive
12h 24h 36h 48h … several days
peak
Eyewitness
Outsiders
Instantaneous
Temporal Distribution: Sources
12h 24h 36h 48h … several days
peak
Eyewitness
Outsiders
Media
Instantaneous
Temporal Distribution: Sources
12h 24h 36h 48h … several days
peak
Eyewitness
Government
Outsiders
Media
Instantaneous
Temporal Distribution: Sources
12h 24h 36h 48h … several days
peak
Eyewitness
Government
NGOs
Outsiders
Media
Instantaneous
Temporal Distribution: Sources
12h 24h 36h 48h … several days
peak
Eyewitness
Government
Business
NGOs
Outsiders
Media
Instantaneous
Temporal Distribution: Sources
Take-AwayTwitter is a medium through which the nuance of events is amplified; yet, when looking at the same data at a higher-
level we see commonalities and patterns. !!
☞ Download all our collections from crisislex.org
Thanks!
Karl Aberer
@ChatoX
@velofemme
Patrick Meier
Questions?
@o_saja
Temporal Distribution ProgressiveInstantaneous
• Start: when the event occurs • High volumes of tweets right
after onset
• Start: when the hazard is detected • High volumes around the peak of
the event (e.g., affects a densely populated area, high economic damage)
Goal: Retrieve comprehensive collections of event-related messages
• Keyword-based sampling: #sandy, #bostonbombings, #qldflood
• Location-based sampling: tweets geo-tagged in disaster area
Problem: Create event collections without too many off-topic tweets
Recall(% of on-crisis tweets retrieved)
Prec
isio
n (o
n-cr
isis
% o
f ret
rieve
d tw
eets
)
DesiredKW-based
Geo-basedLexicon-based
ICWSM 2014, Olteanu et al.
Efficient Data Collection
CrisisLex
API limits • rigid query language • limited volumes Laconic language
Challenges
damage affected people people displaced donate blood text redcross stay safe crisis deepens evacuated toll raises ……
ICWSM 2014, Olteanu et al.
Association-Rules
☞ Diffused events have more than average caution & advice messages ☞ valid for 24/26 events
!☞ Human-induced & accidental events have less than
average eyewitness accounts ☞ valid for 21/26 events
Informativeness & Content Redundancy
☞ Informativeness ☞ Crisis-related: 89% on average (min. 64%, max. 100%) ☞ Informative (from crisis-related): 69% on average (min.
44%, max. 92%) !
☞ Redundancy ☞ NGOs & Government ☞ top 3 messages account for 20%-22% of messages
!☞ Caution & Advice and Infrastructure & Utilities ☞ top 3 messages account for 12%-14% of messages