Transcript Data Mining and Machine Learning Lab

Beyond Crowdsourcing for HADR
Huan Liu, Shamanth Kumar and Huiji Gao
Data Mining and Machine Learning Lab
Outline
‒ Motivation
– Crowdsourcing for Disaster Relief
– Inadequacies of Current Crowdsourcing Systems
– Our Methodology
– Demonstration
Data Mining and Machine Learning Lab
Motivation
• Catastrophic Disasters:
 Haiti earthquake/cholera.
 Middle–east revolutions.
 Japanese tsunami and earthquake.
• Social media for disaster relief:
 Revolutionize the role of media
 Information disseminator
 Communication tool
Data Mining and Machine Learning Lab
Social Media for Crowdsourcing
• Crowdsourcing leverages participatory social media
services and tools to collect information
• Crowdsourcing allows capable crowds to participate in
various HADR tasks.
• Crowdsourcing integrated with crisis map has become a
powerful tool in humanitarian assistance and disaster
relief (HADR).
Data Mining and Machine Learning Lab
Applications of Social Media & Crowdsourcing
for Disaster Relief
• Uses for Individuals
–
–
–
–
–
Find missing people
Early warning of disasters
Get information on relief work progress
Find location of shelters & medical resources
Get in touch with officials and relief workers (more ways to ask for
help)
• Uses for Agencies
–
–
–
–
–
Get situational awareness first hand from citizen reporters
Coordination platform
Send updates on progress of relief work
Discredit rumors
Obtain public feedback
Data Mining and Machine Learning Lab
Inadequacies of Current Crowdsourcing Systems
• Information is hidden in massive and noisy data
– Numerous social media sources
– Unfiltered information can be hard to interpret
– Too many messages can be overwhelming for intelligent decision making
• Lack of a common coordination mechanism
– Different focus and capabilities of HADR agencies
– Hard to optimize resource allocation and distribution
Data Mining and Machine Learning Lab
How We Can Help
• Building crowdsourcing systems to aid in event analysis
– Automate data collection & data storage for event analysis
– Preprocessing and summarize collected data for quick interpretation
– Visualize crowdsourced data
• Building a coordination system for better collaboration
– Coordination mechanism designed for disaster relief
– Intelligent crisis map view to facilitate the response
– Enhancing communications among agencies
Data Mining and Machine Learning Lab
Our tools
ACT
BlogTrackers
TweetTrackers
 Crowdsourced information
 Crowdsourced information
 Crowdsourced information
 Groupsourced information
 Feedback information source
 Situational awareness
 Multi-layer requests view
 Situational awareness
 Near real time information
 Inter agency coordination
 Post event analysis
aggregation
 Post event analysis
Data Mining and Machine Learning Lab
ACT (ASU Coordination Tracker)
Four Modules
•
•
•
•
Request Collection
-Crowdsourcing
-Groupsourcing
Response
Coordination
Statistics
User
User
CrowdSourcing
Reports
Cooperation
Data Mining
Organization
Response
Organization
Response
User
User
User
Cooperation
User
Requests
Pool
Response
Organization
Collector
Collector
ASU Event Map
Response
Response
GroupSourcing
Statistics
Cooperation
Collector Collector Collector
Organization
Data Mining and Machine Learning Lab
Cooperation
Organization
BlogTrackers
Traffic Pattern
Analysis
Three modules
Blog Analysis
• Data Collection
RSS Crawler
(Scheduled Crawling)
• Crowdsourcing
• Analysis Module
• Visualization
Organization
Feedback
Bloggers
Blogosphere
Stored
Blogposts
Bloggers
Batch Crawler
(Bulk Crawling)
Situational Awareness
Blogger Analysis
Influential
Bloggers Analysis
Organization
Data Mining and Machine Learning Lab
TweetTrackers
Three modules
• Data Collection
– Crowdsourcing
• Analysis
• Visualization
Data Mining and Machine Learning Lab
Acknowledgments
• DMML members, in particular, Geoff Barbier,
Fred Morstatter, and Patrick Mcinerney.
• This work benefits from the ONR’s vision on
Social Computing, Digital Revolution, and
HA/DR.
Office of Naval Research
Data Mining and Machine Learning Lab