BIG CRISIS DATA An Open Invitation CARLOS CASTILLO @BigCrisisData Manaus, Brasil, Outubro 2015
BIG CRISIS DATAAn Open Invitation
CARLOS CASTILLO@BigCrisisData
Manaus, Brasil, Outubro 2015
Big Crisis Data — Carlos Castillo 2
This talk is about ...● Disasters and time-critical situations
– Natural, social, or technological hazards
– Mass convergence events● Social media
– Particularly microtext● Computing
– Applications of many fields including NLP, ML, IR
Big Crisis Data — Carlos Castillo 3
http://www.youtube.com/watch?v=0UFsJhYBxzY
Big Crisis Data — Carlos Castillo 4
An earthquake hits a Twitter user
http://xkcd.com/723/
● When an earthquake strikes, the first tweets are posted 20-30 seconds later
● Damaging seismic waves travel at 3-5 km/s, while network communications are light speed on fiber/copper + latency
● After ~100km seismic waves may be overtaken by tweets about them
Big Crisis Data — Carlos Castillo 5
January 2010
How/when did it start for me?
Big Crisis Data — Carlos Castillo 6
Humanitarian Computing
At least 775 publications:
● Crisis Analysis (55)
● Crisis Management (309)
● Situational Awareness (67)
● Social Media (231)
● Mobile Phones (74)
● Crowdsourcing (116)
● Software and Tools (97)
● Human-Computer Interaction (28)
● Natural Language Processing (33)
● Trust and Security (33)
● Geographical Analysis (53)
Source: http://humanitariancomp.referata.com/
Big Crisis Data — Carlos Castillo 7
Humanitarian Computing Topics
Big Crisis Data — Carlos Castillo 8
Big Crisis Data — Carlos Castillo 9
Big Crisis Data — Carlos Castillo 10
Fertile grounds for applied research✔ Problems of global significance
✔ Solved with labor-intensive methods
✔ Better solution provides a public good
✔ Large and noisy data sets available
✔ Engage volunteer communities
Big Crisis Data — Carlos Castillo 11
Fertile grounds for applied research✔ Problems of global significance
✔ Solved with labor-intensive methods
✔ Better solution provides a public good
✔ Large and noisy data sets available
✔ Engage volunteer communities
Relevance to practitioners?
Big Crisis Data — Carlos Castillo 12
Recent collaboratorsPatrick Meier
Sarah Vieweg– QCRI
Muhammad Imran– QCRI
Irina Temnikova– QCRI
Alexandra Olteanu– EPFL
Aditi Gupta– IIIT Delhi
“P.K.” Kumaraguru– IIIT Delhi
Fernando Diaz– Microsoft
Big Crisis Data — Carlos Castillo 13
Outline
Volume
Vagueness
Visualization
Volunteering
Values
Big Crisis Data — Carlos Castillo 14
Disaster Communications
and Scale
Big Crisis Data — Carlos Castillo 15
Crises and disasters● Crises are unstable situations
– May or may not lead to a disaster● Disasters are social phenomena
– Disruptions of routines
Big Crisis Data — Carlos Castillo 16
Temporal and Spatial Dimensions
Big Crisis Data — Carlos Castillo 17
Examples
Big Crisis Data — Carlos Castillo 18
REEL LIFE OR REAL LIFE?
Big Crisis Data — Carlos Castillo 19
REEL LIFE OR REAL LIFE?
Big Crisis Data — Carlos Castillo 20
https://www.youtube.com/watch?v=MylI8HmgMBk
Big Crisis Data — Carlos Castillo 21
In Real Life ...● Some people panic, most people don't
● People gather information from familiar sources
● People quickly decide whether to flee, take cover, or take action
● People improvise complex rescue operations on the spot
Devon, UK, June 2014 London, UK, May 2015 San José Boquerón, Paraguay, Oct 2013
Big Crisis Data — Carlos Castillo 22
Example Disaster-Related Messages“OMG! The fire seems out of control: It’s running down the hills!”
Bush fire near Marseilles, France, in 2009 [Longueville et al. 2009]
“Red River at East Grand Forks is 48.70 feet, +20.7 feet of flood stage, -5.65 feet of 1997 crest. #flood09”
Red River Valley floods in 2009 [Starbird et al. 2010]
“My moms backyard in Hatteras. That dock is usually about 3 feet above water [photo]”
Hurricane Sandy 2013 [Leavitt and Clark 2014]
“Sirens going off now!! Take cover...be safe!”Moore Tornado 2013 [Blanford et al. 2014].
“There is shooting at Utøya, my little sister is there and just called home!”
2011 attacks in Norway [Perng et al. 2013]
Big Crisis Data — Carlos Castillo 23
Social media usage during disasters● Interpersonal (horizontal)
– Stay in touch with family and friends● Citizen sensing (bottom-up)
– Read/Write reports on ground situation● Official communications (top-down)
– E.g. advice, warnings, or evacuation orders
Big Crisis Data — Carlos Castillo 24
Scale: Tweets per Second
Big Crisis Data — Carlos Castillo 25
Requirements● Typical users
– Emergency response services
– Humanitarian relief agencies
– Journalists and the Public● Underspecified requirements that vary over time
● Usually a combination of:
1) Capture the “Big Picture”
2) Obtain “Actionable Insights”
Big Crisis Data — Carlos Castillo 26
Understanding, Classifying and
Extracting
Big Crisis Data — Carlos Castillo 27
Example
“Media must report about d alleged 20k RSS chaps off 2 #Nepal.here’s a pic coz d 1 @ShainaNC shared isn’t true.. ;)”
Big Crisis Data — Carlos Castillo 28
Social media messages● Social media is more like a transcript of a conversation than like
text meant to stand on its own
– Awkward entry methods:● Fragmented language and incomplete sentences● Many typographic and grammatical errors
– Conversational:● Little or no context (hard to comprehend in isolation)● Code switching and borrowing● Internet slang
Big Crisis Data — Carlos Castillo 29
Slang
Big Crisis Data — Carlos Castillo 30
ClassificationCaution &
AdviceInformation
SourcesDamage &Casualties Donations
Gov
Eyewitness
Media
NGO
Outsider
...
...
Filteredtweets
Big Crisis Data — Carlos Castillo 31
Classification Axes● By usefulness (application-dependent!)
– Not related, Related but useless, Useful● By factual, subjective, or emotional content
● By information provided
● By information source
– Government, NGOs, media, eyewitnesses, etc.● By humanitarian clusters
Big Crisis Data — Carlos Castillo 32
Humanitarian Clusters
Big Crisis Data — Carlos Castillo 33
Alexandra Olteanu, Sarah Vieweg and Carlos Castillo: What to Expect When the Unexpected Happens: Social Media Communications Across Crises.To appear in CSCW 2015.
Humanitarian Clusters (cont.)
Big Crisis Data — Carlos Castillo 34
A large-scale study of crisis tweets● Collect tweets from 26 disasters
● Classify according to:
● Informative / Not informative● Information provided● Information source
● Several iterations required to write the “right” instructions
Alexandra Olteanu, Sarah Vieweg and Carlos Castillo: "What to Expect When the Unexpected Happens: Social Media Communications Across Crises" In CSCW 2015, 14-18 March in Vancouver, Canada. ACM Press.
Big Crisis Data — Carlos Castillo 35
Information Provided in Crisis Tweets
N=26; Data available at http://crisislex.org/
Big Crisis Data — Carlos Castillo 36
What do people tweet about?● Affected individuals
– 20% on average (min. 5%, max. 57%)
– most prevalent in human-induced, focalized & instantaneous events
● Sympathy and emotional support
– 20% on average (min. 3%, max. 52%)
– most prevalent in instantaneous events● Other useful information
– 32% on average (min. 7%, max. 59%)
– least prevalent in diffused events
Big Crisis Data — Carlos Castillo 37
What do people tweet about? (cont.)● Infrastructure and utilities
– 7% on average (min. 0%, max. 22%)
– most prevalent in diffused events, in particular floods● Caution and advice
– 10% on average (min. 0%, max. 34%)
– least prevalent in instantaneous & human-induced events● Donations and volunteering
– 10% on average (min. 0%, max. 44%)
– most prevalent in natural hazards
Big Crisis Data — Carlos Castillo 38
Distribution over information sources
Big Crisis Data — Carlos Castillo 39
Distribution over time
Big Crisis Data — Carlos Castillo 40
Dataset
CrisisLexT26
www.crisislex.org
Big Crisis Data — Carlos Castillo 41
Information Extraction
...
Classifiedtweets @JimFreund: Apparently we have no choice.
There is a tornado watch in effect
tonight.
Big Crisis Data — Carlos Castillo 42
Extraction● #hashtags, @user mentions, URLs, etc.
– Regular expressions
– Text library from Twitter● Temporal expressions
– Part-of-speech tagger + heuristics
– Natty library● Supervised learning
Muhammad Imran, Shady Elbassuoni, Carlos Castillo, Fernando Diaz and Patrick Meier: Practical Extraction of Disaster-Relevant Information from Social Media. Social Web and Disaster Management (SWDM) workshop. Rio de Janeiro, Brazil, 2013.
Big Crisis Data — Carlos Castillo 43
Labels for extraction● Type-dependent instruction
● Ask evaluators to copy-paste a word/phrase from each tweet
Big Crisis Data — Carlos Castillo 44
Learning: Conditional Random Fields
● Extends HMM to incorporate more possible dependencies
● Used extensively in NLP for part-of-speech tagging and information extraction
HMM Linear-chain CRF
hidden
observed
Big Crisis Data — Carlos Castillo 45
Tool● CMU ARK Twitter NLP
– Tokenization
– Feature extraction
– CRF learning● Very easy to use
– simply change the training set (part-of-speech tags),
– then re-train
Big Crisis Data — Carlos Castillo 46
Output examplesRT @weatherchannel: .@NYGovCuomo orders closing of NYC bridges. Only Staten Island bridges unaffected at this time. Bridges must close by 7pm. #Sandy #NYC
Wow what a mess #Sandy has made. Be sure to check on the elderly and homeless please! Thoughts and prayers to all affected
RT @twc_hurricane: Wind gusts over 60 mph are being reported at Central Park and JFK airport in #NYC this hour. #Sandy
RT @mitchellreports: Red Cross tells us grateful for Romney donation but prefer people send money or donate blood dont collect goods NOT best way to help #Sandy
Big Crisis Data — Carlos Castillo 47
Extractor evaluation
Setting Rec Prec
Train 2/3 Joplin, Test 1/3 Joplin 78% 90%
Train 2/3 Sandy, Test 1/3 Sandy 41% 79%
Train Joplin, Test Sandy 11% 78%
Train Joplin + 10% Sandy, Test 90% Sandy
21% 81%
● Precision is: one word or more in common with what humans extracted
Big Crisis Data — Carlos Castillo 48
Donations matching● Identify and match requests/offers for donations
– Money, clothing, food, shelter, volunteers, blood● Method
– Classify
– Determine key aspects
– Extract key aspects
– Per-aspect matching
Hemant Purohit, Amit Sheth, Carlos Castillo, Patrick Meier, Fernando Diaz: Emergency-Relief Coordination on Social Media: Automatically Matching Resource Requests and Offers. First Monday 19 (1), January 2014.
Big Crisis Data — Carlos Castillo 49
Donations matching
Average precision = 0.21 (0.16 if only text similarity is used)
Big Crisis Data — Carlos Castillo 50
Crisis maps from social
media
Big Crisis Data — Carlos Castillo 51
Big Crisis Data — Carlos Castillo 52
Big Crisis Data — Carlos Castillo 53
Patrick Meier, Social Innovation Director @ QCRI – http://irevolution.net/
“What can speed humanitarian
response to tsunami-ravaged coasts?
Expose human rights atrocities?
Launch helicopters to rescue
earthquake victims? Outwit corrupt
regimes?
A map.”
Big Crisis Data — Carlos Castillo 54
Crisis mapping goes mainstream (2011)
Big Crisis Data — Carlos Castillo 55
Big Crisis Data — Carlos Castillo 56
Big Crisis Data — Carlos Castillo 57
Big Crisis Data — Carlos Castillo 58
Big Crisis Data — Carlos Castillo 59
Big Crisis Data — Carlos Castillo 60
Automatic Mapping (floods)● Top: hydrological data
● Bottom: tweet density
● Broad match with affected areas
● Many biases towards places with higher density of smartphones
De Albuquerque, João Porto, Herfort, Benjamin, Brenning, Alexander, and Zipf, Alexander. 2015. A geographic approach for combining social media and authoritative data towards identifying useful information for disaster management. International Journal of Geographical Information Science, 29(4), 667–689.
Big Crisis Data — Carlos Castillo 61
Automatic Mapping (Dengue)
Gomide, Janaina and Veloso, Adriano and Meira, Wagner and Almeida, Virgilio and Benevenuto, Fabricio and Ferraz, Fernanda and Teixeira, Mauro (2011) Dengue surveillance based on a computational model of spatio-temporal locality of Twitter. pp. 1-8. In: Proceedings of the ACM WebSci'11, June 14-17 2011, Koblenz, Germany.
● Top: official reports
● Bottom: tweets
Big Crisis Data — Carlos Castillo 62
Current Approach
Hybrid real-time systems
MicroMappers
Manual processing: crowdsourcing
Automatic processing: machine learning
Big Crisis Data — Carlos Castillo 63
http://newsbeatsocial.com/watch/0_s6xxcr3p
Big Crisis Data — Carlos Castillo 64
Big Crisis Data — Carlos Castillo 65
Big Crisis Data — Carlos Castillo 66
https://www.youtube.com/watch?v=uKgE3yWJ0_I
Big Crisis Data — Carlos Castillo 67
Volunteering and Values
Big Crisis Data — Carlos Castillo 68
Volunteering is a constant● Integral part of how communities react to disasters
● Organizational types:
– Existing – Extending – Expanding – Emerging● Emergent organizations a mixed blessing for existing ones
● New scenario: digital volunteering
– E.g. volunteer annotations, including crisis mapping
Big Crisis Data — Carlos Castillo 69
Why do people volunteer?
Altruism is key, but it's
one of many reasons
Big Crisis Data — Carlos Castillo 70
Privacy and Ethics● Protect the privacy of individuals
– ICRC Data Protection Guidelines
– UN Guidelines on Cyber Security● Protect victims and responders during armed attacks
● Protect volunteers from distal exposure
● Protect citizen reporters from danger and retaliation
● Give back and share results and data
Big Crisis Data — Carlos Castillo 71
“I'm dying, they are tweeting”
Digital Voyeurism
Big Crisis Data — Carlos Castillo 72
CONCLUSIONS
Computationally feasible
Supported by
data
Useful
Good projects in this space
Computationally feasible
Supported by
data
Useful
Good projects in this space
Temptation! Danger!
Poorly planned projects :-(
AI-complete problems
Big Crisis Data — Carlos Castillo 75
Interdisciplinary Research● As many things, it has Good, Bad, and Ugly aspects● Good
– You learn a lot, and it's the only way of supporting claims of practical utility in applied research
● Bad– Formal response organizations can be very difficult to engage with;
relationships should be established between operations● Ugly
– Working software and 24/7 support for a critical need now vs advanced proof-of-concept later
Possibility of large impact by using computer science to support
humanitarian work
=Applied computing at its best
Big Crisis Data — Carlos Castillo 77
References● Carlos Castillo: “Big Crisis Data.” Cambridge University Press, 2016 (forthcoming).● Muhammad Imran, Carlos Castillo, Fernando Diaz, Sarah Vieweg: "Processing Social Media Messages in Mass
Emergency: A Survey" In ACM Computing Surveys, Volume 47, Issue 4, June 2015.● Alexandra Olteanu, Sarah Vieweg and Carlos Castillo: "What to Expect When the Unexpected Happens: Social
Media Communications Across Crises" In CSCW 2015, 14-18 March in Vancouver, Canada. ACM Press. ● Muhammad Imran, Ioanna Lykourentzou, Yannick Naudet and Carlos Castillo: Engineering Crowdsourced Stream
Processing Systems. Technical report, 2015.● Hemant Purohit, Amit Sheth, Carlos Castillo, Patrick Meier, Fernando Diaz: Emergency-Relief Coordination on
Social Media: Automatically Matching Resource Requests and Offers. First Monday 19 (1), January 2014. ● Sarah Vieweg, Carlos Castillo and Muhammad Imran: "Integrating Social Media Communications into the Rapid
Assessment of Sudden Onset Disasters." SocInfo 2014.● Alexandra Olteanu, Carlos Castillo, Fernando Diaz and Sarah Vieweg: CrisisLex: A Lexicon for Collecting and
Filtering Microblogged Communications in Crises. In ICWSM. Ann Arbor, MI, USA. June 2014. ● Carlos Castillo, Marcelo Mendoza, Barbara Poblete: Predicting Information Credibility in Time-Sensitive Social
Media (+Supplementary Material). In Internet Research, Vol. 23, Issue 5, Special issue on The Predictive Power of Social Media, pp. 560-588. October 2013.
● Muhammad Imran, Shady Elbassuoni, Carlos Castillo, Fernando Diaz and Patrick Meier: Practical Extraction of Disaster-Relevant Information from Social Media. Social Web and Disaster Management (SWDM) workshop. Rio de Janeiro, Brazil, 2013.
● Muhammad Imran, Shady Elbassuoni, Carlos Castillo, Fernando Diaz and Patrick Meier: Extracting Information Nuggets from Disaster-Related Messages in Social Media. In ISCRAM. Baden-Baden, Germany, 2013. Best paper award.
Big Crisis Data — Carlos Castillo 78
Thank you!Follow @BigCrisisData