This document presents an approach to infer resource needs from tweets during crisis situations using a domain ontology. It filters tweets by location and identifies needs related to power, medical care, food, and water. Text-based location detection identified locations for 66% of tweets within the affected region, compared to 43% for metadata-based detection. The approach uses DBpedia's ontology to precisely detect locations mentioned in text. This information could help coordinate response by identifying critical resource needs and their locations.

1. Assisting Coordination during Crisis: A
Domain Ontology based Approach to Infer
Resource Needs from Tweets
Shreyansh Bhatt, Hemant Purohit, Andrew Hampton, Valerie Shalin,
Amit Sheth, John Flach

2. 2
• Community responds to the scale of disaster
• (Informal) Social media can be leveraged to help coordination
Variety of responses on social media!

3. 3
Challenges
Filter
• High Volume and
Velocity:
• 20M+ Tweets during
1st week of #Sandy
• Identify resources
Semantics
• Citizen do not
always write about
specific needs
Locate Resources
• Metadata location
(device sensor &
user profile)
• ~ 21% tweets with
metadata location
Domain
model :
Crisis
Ontology
Technique
to precisely
identify text
location
Potential
Solutions

4. 4
Approach

5. 5
Location Detection
• Two fold filtering to increase precision:
• 1.) Named Entity based, and 2.) DBpedia ontology based
• Use of DBpedia allows annotation of city, state and famous places

6. 6
Medical/power related tweets
• Medical emergency due to power cut at hospitals in Brooklyn was
identified by power cut information (News: http://j.mp/2Hospitals)
Time
(2012)
Message text (Power related) Text Location
Identified
Oct. 29,
21:30:02
Lots of wind and some rain but still running and
no power outage in Clinton Hill, Brooklyn. #Sandy
#Hurricane
http://dbpedia.org/reso
urce/Brooklyn
Oct. 29,
23:56:57
Power cut to coney island and Brighton beach
#HurricaneSandy #NYC
http://dbpedia.org/reso
urce/Coney_Island
Oct. 30,
01:30:36
Power may be cut off soon in south bklyn. Coney,
Gravesend Sheedshed Bay etc #Sandy
#Frankenstorm
http://dbpedia.org/reso
urce/Coney_Island

7. 7
Medical/power related tweets (cont.)
• Medical emergency due to power cut at hospitals in Brooklyn was
identified by power cut information (News: http://j.mp/2Hospitals)
Time
(2012)
Message text (Medical related) Text Location
Identified
Oct. 30,
03:15:08
@911BUFF: BREAKING CONEY ISLAND
HOSPITAL ON FIRE. NYU HOSP. EVACUATED,
BELLEVUE HOSPITAL ALSO LOSING BACKUP
POWER #SANDY #NYC #frankenstorm
http://dbpedia.org/resourc
e/Bellevue_Hospital_Cent
er
Oct. 30,
20:20:42
SANDY: Bellevue Hospital is on backup power,
trying to evacuate as much as possible, 2 young
boys missing from SI since beginning of
Hurricane
http://dbpedia.org/resourc
e/Bellevue_Hospital_Cent
er

8. 8
Tweets about resources with location types
• Ability to infer location from text increases location information over
tweet metadata information by approximately 50%
Total
Text
location
Metadata
location
Text w/o
Metadata
Text and
Metadata
Metadata
w/o text
Power 103102 14969 24361 10974 3995 20366
Medical 16002 3243 6015 2057 1186 4829
Food/water 38952 5046 9152 3574 1472 7680
Power &
Medical
948 231 377 134 97 280
Power &
Food
2908 382 839 260 122 717
Power &
food &
medical
44 39 30 13 26 17

9. 9
Text location vs Metadata location
• Text-location detection precision : 88%
• 66% text-locations within affected region of disaster
Location
source
Total
tweets
Total tweets with
location in affected
region
Total tweets with
location not in
affected region
Text 517 340 (66%) 177 (34%)
Meta-data 313 238 (43%) 323 (57%)

10. 10
Summary
Visit our poster for More insights!
• Domain knowledge based approach to identify contextually interdependent
resource needs reported via social media
• DBpedia ontology driven approach for precise text-location detection
• Next Steps
• Exploit behaviors of seeking-supplying of resources in messages
• Improve location information verification
– Ack: NSF SoCS project, grant IIS-1111182, ‘Social Media Enhanced
Organizational Sensemaking in Emergency Response’
• http://www.knoesis.org/research/semsoc/projects/socs
• Questions?
• Mail: {shreyansh,hemant}@knoesis.org, Tweet: @hemant_pt