Mobile Phone Enabled Social Community Extraction for ... · Location - Theory, Practice and...
Transcript of Mobile Phone Enabled Social Community Extraction for ... · Location - Theory, Practice and...
Mobile Phone Enabled Social Community
Extraction for Controlling of Disease
Propagation in Healthcare
Yingying (Jennifer) Chen
Director of Data Analysis and Information Security Lab
Graduate Director of Networked Information Systems (NIS)
Department of Electrical and Computer Engineering
Stevens Institute of Technology
Hoboken, NJ, USA
Email: [email protected]
Web: personal.stevens.edu/~ychen6
March 2014
DAISY Data Analysis and Information SecuritY Lab
Research Interests
Cyber security and privacy
Secure localization systems; privacy in Location Based Services (LBS)
Security in mobile cloud computing, smartphone privacy
Wireless security: identity-based attacks, anti-jamming, intrusion detection
Security in smart grid and cognitive radio networks
Mobile healthcare and social networks Wellbeing monitoring; smart vaccine delivery
User authentication
Activity recognition
Wireless localization and context-aware sensing Wireless localization algorithms & testbeds (using mobile phone, sensor, RFID)
Personnel and asset monitoring and tracking
Smartphone enabled mobile social networks
Mobile Computing and vehicular applications Detect driver phone use to reduce driver distraction caused by cell phone usage
Exploit cell phone signal strength to perform vehicular speed estimation
2
Data Analysis and Information Security Lab
(DAISY): www.stevens.edu/daisy
Funding
National Science Foundation, Google Research, Department of Defense (DoD), Army Research Office, and Air Force Research Labs
Collaborators
Industry: AT&T, IBM T. J. Watson Research Center, InfoBeyond.
Academia: Columbia University, Rutgers University, Robert Wood Medical School, Lehigh University, University of South Carolina.
Students
Postdoc researchers, Ph.D., master and undergraduate students
3
Publications and Awards
Book: Yingying Chen, Wenyuan Xu, Wade Trappe, Yanyong Zhang, Securing Emerging Wireless
Systems, ISBN:978-0-387-88490-5,Springer, 2009.
Book chapters: Jie Yang, Yingying Chen, Richard P. Martin, Wade Trappe and Marco Gruteser, "On the
Performance of Wireless Indoor Localization Using Received Signal Strength", Position Location - Theory, Practice and Advances: A Handbook for Engineers and Academics, Seyed A. (Reza) Zekavat, Mike Buehrer (eds.), Wiley-IEEE, chapter 12, 2011.
Yingying Chen, Jie Yang, Wade Trappe and Richard P. Martin, "Impact of Anchor Placement and Anchor Selection on Localization Accuracy", Position Location - Theory, Practice and Advances: A Handbook for Engineers and Academics, Seyed A. (Reza) Zekavat, Mike Buehrer (eds.), Wiley-IEEE, chapter 13, 2011.
Yingying Chen, Jie Yang, Wade Trappe and Richard P. Martin, "Defending against Identity-Based Attacks in Wireless Networks", Securing Cyber-Physical Infrastructures: Foundations and Challenges, Sajal Das, Krishna Kant and Nan Zhang (eds.), chapter 4, 2011.
Published over 80 journal articles and referred conference papers.
National Science Foundation CAREER Award 2010.
Google Faculty Research Award 2010.
IEEE Outstanding Contribution Award from IEEE New Jersey Coast Section, 2005, 2006, 2007, 2008 and 2009.
Best Paper Award of the ACM International Conference on Mobile Computing and Networking (ACM MobiCom) 2011.
4
5
Talk Outline
Introduction and Motivation
Framework Overview
Disease Propagation Model
Community and Kernel Structure Extraction
Vaccine and Alert Messages Distribution
Simulation Evaluation
Conclusion
6
Introduction and Motivation
Rapid deployment of sensing technology in mobile
phones.
The collected sensing data can be used for supporting a
broad range of applications:
monitoring/tracking, first responder and military applications.
Few studies used mobile phone-based sensor data to
address healthcare related problem.
7
Framework Overview
Main components
Disease propagation model.
Community and kernel structure extraction.
Vaccine and alert messages distribution.
8
Disease Propagation Model
Four states are used grounded on the standard
epidemic SIR model:
Susceptible without alert;
Susceptible with alert;
Infective;
Immunized;
9
Community and Kernel Structure Extraction
Main components:
Flow overview Dividing encounter events into multiple trace to construct contact graph
and extract communities and kernel structures.
Merging the extracted communities and kernel structures with existing
community and kernel structures.
10
Mobile Device Trace Segmentation
Each entry of traces includes the starting and
ending time of the contact as well as IDs of the
mobile devices.
Assume that we have D time windows:
Divide the contact trace into multiple trace pieces
according to these time windows.
0 1 1 2 1[ , ],[ , ],...,[ , ]D DT T T T T T-
11
Construct Contact Graphs
Contact graph G=(V,E) which consists of a
vertex set V and an edge set E.
Each vertex denotes a person and each edge
denotes two persons encounter for at least W
times.
We choose for community and
for kernel structure with .
1W w=2W w=
2 1w w>
12
Community Extraction and Merging
Communities are extracted from contact
graph by hierarchical clustering and modularity
Q in time window i:
Merging the communities using the community
merge operation for two communities if they
satisfy:
j
iA
13
Vaccine and Alert Message Distribution
Two types of messages received by users:
Vaccination messages: a user should go to obtain a
vaccine shot;
Alert Messages: a user should take precautions as
directed.
The kernel structures are considered first:
The top K users are selected for vaccination or alert
messages;
The weight of each person reflects the priority;
kV
14
Simulation Evaluation
Traces collected from smart phones:
MIT traces (97 participants, 20 days)
Implemented our framework in a home-grown simulator;
Compare with existing techniques:
Random Distribution method: server randomly chooses the
users to receive the vaccination and alert messages.
Encounter-based method: vaccination and alert messages
are sent if sick person encounters with a susceptible
person.
Simulation setup:
Extensive experiments on MIT trace by varying different
parameters in our epidemiology infection model.
15
Simulation Results: Ratio of Infected Persons (1)
Ratio of infected person Versus Different recovery cycle
Our proposed method achieves a lower infection ratio than Random
Distribution and Encounter-based methods for each recovery
cycle.
16
Simulation Results: Ratio of Infected Persons (2)
Our proposed method achieves a lower final infection ratio than
other methods.
17
Conclusion
Explored the design of a mobile phone enabled
community based disease propagation rate reduction
framework for the healthcare domain.
Developed an online algorithm for identifying dynamic
communities of different relationship levels .
Designed a community-based framework to make
decisions on the alert messages and vaccine shot
distribution.
Simulation studies show our community based disease
control scheme results in lower final infection ratio than
existing techniques
18
Thanks and Questions ?