ITU Kaleidoscope 2015
Trust in the Informa0on Society
Network Failure Detection System
for Traffic Control
using Social Information
in Large-Scale Disasters
Chihiro Maru
1
, Miki Enoki
1,2
, Nakao Akihiro
3
,
Shu Yamamoto
3
, Saneyasu Yamaguchi
4
, Oguchi Masato
1
1
Ochanomizu University,
2
IBM Research – Tokyo,
3
University of Tokyo,
4
Kogakuin University
Outline
p
Background
p
Problem
p
Objective
p
Novelty
p
Network Failure Detection System
–
Telephony Failure Detection
–
Location-aware Prioritization
p
Network Control System
Overview
of the Great East Japan Earthquake
p
Occurred at 14:46 JST on 11 March 2011
p
Magnitude 9.0
p
15,893 dead and 6,152 injured
p
Wiped out by “Tsunami” (huge wave)
Barcelona, Spain, 9-11 December 2015
ITU Kaleidoscope 2015 - Trust in the Information Society
3
Otsuchi town
http://www.47news.jp/photo/
http://www.tenki.jp/
Miyagi
Tokyo
Problem
p
Difficult to grasp
all network condition
immediately using only information from
Role of SNS (e.g., Twitter)
p
Users care network condition and provide
such information through
social networking
service (SNS)
p
is one of the most widely used SNS’s
Barcelona, Spain, 9-11 December 2015
ITU Kaleidoscope 2015 - Trust in the Information Society
5
I cannot
get through
to Tokyo
I’m in Kyoto
now. I cannot
get through.
Objective
Achieve
automatic/autonomic network control
utilizing
collective knowledge
Novelty of this work
p
We perform the network control using the
collective intelligence of social networking
service
p
ITU-T Focus Group on Disaster Relief
Systems performs the network control
using
wireless sensor networks
p
Our approach is complementary to ITU-T
Focus Group on Disaster Relief Systems
Barcelona, Spain, 9-11 December 2015
Network Failure Detection System
Earthquake Early
Warning (EEW)
Network Failure Detec7on System
Output Data
Location
Aware
Classification
Candidate Data
Detection
Tweet
Collection
of Tweets
Network Control
System
Calculation
of
Rate of
Importance
Keyword Search
Barcelona, Spain, 9-11 December 2015
ITU Kaleidoscope 2015 - Trust in the Information Society
Earthquake Early
Warning (EEW)
Network Failure Detec7on System
Output Data
Location
Aware
Classification
Candidate Data
Detection
Tweet
Collection
of Tweets
Network Control
System
9
Calculation
of
Rate of
Importance
Ø
Set specific keywords that
represent telephone trouble
Candidate Data Detection Method
Tweets of
Keyword Search
Loca7on
Informa7on
Detec7on
Characteris7c
Words
Detec7on
Tweets
Detec7on
Temporal
Filtering
Time Variation of Tweets
p
Cumulative occurrences of tweets exhibit
exponential distributions
p
Determine a
certain threshold
to discard tweets
11
Barcelona, Spain, 9-11 December 2015
ITU Kaleidoscope 2015 - Trust in the Information Society
Time variation of
cumulative frequency is
similar
to the
cumulative distribution
of an exponential
distribution
Temporal Filtering
Ø
Capture 80% of the event
in 60 minutes
Ø
Set the time filter to
60
minutes
is reasonable
Location Aware Classification
Barcelona, Spain, 9-11 December 2015
ITU Kaleidoscope 2015 - Trust in the Information Society
Earthquake Early
Warning (EEW)
Network Failure Detec7on System
Output Data
Location
Aware
Classification
Candidate Data
Detection
Tweet
Collection
of Tweets
Network Control
System
13
Calculation
of
Rate of
Importance
Location Classification
p
Filter out tweets reporting telephony
failures
To LocationA
From LocationB
I’m in
Tokyo
now. I cannot
get through.
I cannot get
through to
Miyagi
…
I’m worried.
Location Classification
p
Filter out tweets
with information on
location associated with failures
Barcelona, Spain, 9-11 December 2015
ITU Kaleidoscope 2015 - Trust in the Information Society
15
To LocationA
From LocationB
I’m in
Tokyo
now. I cannot
get through.
I cannot get
through to
Miyagi
…
I’m worried.
Prioritization
Earthquake Early
Warning (EEW)
Network Failure Detec7on System
Output Data
Location
Aware
Classification
Candidate Data
Detection
Tweet
Collection
of Tweets
Network Control
System
Calculation
of
Rate of
Importance
Calculation of Rate of Importance
about Location Information
p
It has three indicators
1)
Prediction of seismic intensity on the
detected locations
2)
Rate of tweets reporting telephony failures
to the detected locations
3)
Rate of increase of tweets in emergency
p
Prioritize locations for the efficient area
restoration
Barcelona, Spain, 9-11 December 2015
Rate of Importance
in the Great East Japan Earthquake
Ø
Locations with large
damages are predicted
with high priorities
Network Control System
Barcelona, Spain, 9-11 December 2015
ITU Kaleidoscope 2015 - Trust in the Information Society
Earthquake Early
Warning (EEW)
Network Failure Detec7on System
Output Data
Location
Aware
Classification
Candidate Data
Detection
Tweet
Collection
of Tweets
Network Control
System
19
Calculation
of
Rate of
Importance
Network Control System
based on Analysis of SNS
p
Use information on network condition
detected by our system
p
Optimize the network traffic automatically
p
Use an architecture called FLARE[1]
–
Can access data in
application layer
–
Compatible
with our system that try to use
Implementation with
Programmable Nodes : FLARE
p
Integrate Network Failure Detection System
into FLARE Central
Barcelona, Spain, 9-11 December 2015
ITU Kaleidoscope 2015 - Trust in the Information Society
FLARE3
FLARE Central
FLARE Switch
Acquisition and
Analysis
of Social Data
Routing Control
FLARE1
FLARE2
FLARE4
Routing Control based on Information
of Social Networking Service
Before the earthquake
After the earthquake
Contributions of this work
p
Design and Prototype of SNS-based
Network Failure Detection System
–
Detect telephony failures
with a high degree of
accuracy
–
Prioritize locations
using three indicators for
the efficient area restoration
p
Integration on SNS-based Failure
Detection into Network Control
–
Perform the network control automatically
using the collective intelligence of social
networking service
Barcelona, Spain, 9-11 December 2015
References
[1] FLARE, “http://www.pilab.jp/ipop2013/
exhibition/iPOP2013_uTokyo_panel.pdf”
