your friends have more friends than you do: identifying influential mobile users through random...

16
wine.korea.ac .kr WINE Your Friends Have More Friends Than You Do: Identifying Influential Mobile Users Through Random Walks Bo Han, Aravind Srinivasan University of Maryland MobiHoc ‘12 Presented by YongJoon Chung

Upload: rhoda-elliott

Post on 12-Jan-2016

212 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Your Friends Have More Friends Than You Do: Identifying Influential Mobile Users Through Random Walks Bo Han, Aravind Srinivasan University of Maryland

wine.korea.ac.kr

WINE

Your Friends Have More Friends Than You Do: Identifying Influential Mobile Users Through

Random Walks

Bo Han, Aravind SrinivasanUniversity of Maryland

MobiHoc ‘12

Presented by YongJoon Chung

Page 2: Your Friends Have More Friends Than You Do: Identifying Influential Mobile Users Through Random Walks Bo Han, Aravind Srinivasan University of Maryland

wine.korea.ac.kr

WINE 2

Contents

Introduction iWanderControlling Infectious DiseasesFacilitating Information DisseminationConclusion

Page 3: Your Friends Have More Friends Than You Do: Identifying Influential Mobile Users Through Random Walks Bo Han, Aravind Srinivasan University of Maryland

wine.korea.ac.kr

WINE 3

Introduction

Identifying influential users in mobile social net-works.

Previous methods uses centralized algorithms on ei-ther friendship or social-contact graphs of all users

High computational complexity

Proposes iWander, a lightweight and distributed pro-tocol

Extremely low message overhead

Page 4: Your Friends Have More Friends Than You Do: Identifying Influential Mobile Users Through Random Walks Bo Han, Aravind Srinivasan University of Maryland

wine.korea.ac.kr

WINE 4

iWander - Intuition

If a small group of smartphones initialize ran-dom walks periodicaaly, influential mobile users may be visited by these random walks more frequently than average.

Page 5: Your Friends Have More Friends Than You Do: Identifying Influential Mobile Users Through Random Walks Bo Han, Aravind Srinivasan University of Maryland

wine.korea.ac.kr

WINE 5

iWander - Protocol

Every ΔT hours, iWander generates probing message with a given probability q on each smartphone.

Message only contains a pre-configured time-to-live (TTL) filed L. When a smartphone receives a probing message,

it decreases L by 1, stores it in its local queue, and passes the queued message when different peer is in contact. When L is 0, iWander discards the message.

iWander maintains a random-walk counter The higher the counter gets, the more influential the user is.

Page 6: Your Friends Have More Friends Than You Do: Identifying Influential Mobile Users Through Random Walks Bo Han, Aravind Srinivasan University of Maryland

wine.korea.ac.kr

WINE 6

iWander - Protocol

1. Every ΔT hours, senses nearby device2. With probability q, send length T random-walk message

T = 2

• Random walk count +1

• Queue• T = 1

Page 7: Your Friends Have More Friends Than You Do: Identifying Influential Mobile Users Through Random Walks Bo Han, Aravind Srinivasan University of Maryland

wine.korea.ac.kr

WINE 7

Controlling Infectious Diseases

iWander can be used to control infectious dis-eases and perform early outbreak detection.

Vaccinating people with high random-walk counters first can help to contain spread of flu.

Page 8: Your Friends Have More Friends Than You Do: Identifying Influential Mobile Users Through Random Walks Bo Han, Aravind Srinivasan University of Maryland

wine.korea.ac.kr

WINE 8

Controlling Infectious Diseases

Simulation setup C based simulator on the SIR model for spread of infectious

diseases simulation. Used Dartmouth dataset for individual social contact simula-

tion. Dartmouth dataset – records of association and disassociation events

at WiFi access points

Comparison with random immunization.

Page 9: Your Friends Have More Friends Than You Do: Identifying Influential Mobile Users Through Random Walks Bo Han, Aravind Srinivasan University of Maryland

wine.korea.ac.kr

WINE 9

Controlling Infectious Diseases

Page 10: Your Friends Have More Friends Than You Do: Identifying Influential Mobile Users Through Random Walks Bo Han, Aravind Srinivasan University of Maryland

wine.korea.ac.kr

WINE 10

Controlling Infectious Diseases

Page 11: Your Friends Have More Friends Than You Do: Identifying Influential Mobile Users Through Random Walks Bo Han, Aravind Srinivasan University of Maryland

wine.korea.ac.kr

WINE 11

Facilitating Information Dissemination

iWander can be used for target-set selection of information dissemination

Can be used for opportunistic communications and social participation to facilitate information dissem-ination

Use iWander to select target users without requir-ing global network structure

Top k random-walk counters will receive data from server and disseminate to other users.

Page 12: Your Friends Have More Friends Than You Do: Identifying Influential Mobile Users Through Random Walks Bo Han, Aravind Srinivasan University of Maryland

wine.korea.ac.kr

WINE 12

Facilitating Information Dissemination

Simulation setup C based simulator with the same Dartmouth data

set for performance evaluation of random-walk based target-set selection

Compare the performance with random selection. Also compare with hybrid method which uses 10%

from 1-set random walks and 90% from low central-ity.

Page 13: Your Friends Have More Friends Than You Do: Identifying Influential Mobile Users Through Random Walks Bo Han, Aravind Srinivasan University of Maryland

wine.korea.ac.kr

WINE 13

Facilitating Information Dissemination

Page 14: Your Friends Have More Friends Than You Do: Identifying Influential Mobile Users Through Random Walks Bo Han, Aravind Srinivasan University of Maryland

wine.korea.ac.kr

WINE 14

Facilitating Information Dissemination

Page 15: Your Friends Have More Friends Than You Do: Identifying Influential Mobile Users Through Random Walks Bo Han, Aravind Srinivasan University of Maryland

wine.korea.ac.kr

WINE 15

Conclusion

The paper proposed a light weight and dis-tributed protocol for identifying influential mobile users who have high centrality in their social-contact networks.

iWander was evaluated for two applications and outperformed the random approach.

Page 16: Your Friends Have More Friends Than You Do: Identifying Influential Mobile Users Through Random Walks Bo Han, Aravind Srinivasan University of Maryland

wine.korea.ac.kr

WINE 16

Thank you!