2.All-hazardous Disaster Situation Browser

addictedswimmingAI and Robotics

Oct 24, 2013 (3 years and 7 months ago)

52 views

Co
-
PI: Dr.
Shu
-
Ching

Chen,
Professor and Project Lead

Dr
. Tao Li,
Assoc. Professor

Dr
.
Jainendra

K
Navlakha
, Professor

Steven Luis, Project Manager

School of Computing and Information Sciences

Florida International University

1.Business Continuity Information Network

2.All
-
hazardous Disaster Situation Browser


Helping business recover faster


The first web
-
based
Publi
c
-
Private Partnership tool
that
helps the County Business Recovery Program
communicate, share information and collaborate on
disaster events with the private sector


Help the businesses to better asses the impact to the
community and help business find resources to recovery
faster


Provide a gathering place for businesses to report on
available resources (products/services) they can provide


Help County Partners
collect and understand
damage in
the business community immediately following disaster
event


A public service available at
www.bizrecovery.org

and
part of a recognized Public Private Partnershi
p model
by
FEMA Private
S
ector Office




Studies show that ~40% of companies
failed
within 36 months
when they were closed for 3 or
more days as a result of a
hurricane


If BCIN helped
5% of the companies in South
Florida to speed up their hurricane recovery by
one week, it would prevent $220M of non
-
property economic losses which would result
from that week's closure.



Obstacles that hinder business recovery process




Lack of real
-
time, reliable and business relevant information in the aftermath of
a major disaster




Lack of ways of communication and coordination both within the company and
with partner organizations which their business operations depend on




Lack of ways to accurately understand and assess disaster damages to the
infrastructure, organization and operating ecosystem




Lack of ways to reconnect the business supply chain and ecosystem which are
broken or fragmented as the result of the disaster




Lack of tools to assist making disaster recovery decisions encompassing vast
amount of data, multitude of interlocking factors, and sound risk assessment.


What Businesses need: Critical Information



Critical information helps businesses in making key pre
and post storm decisions



What is the status of…?



Key Employees



Facilities



Supply chain and Logistics



Financial Network



Where can I find products/services to help me re
-
open quickly?



When can I re
-
open my business?



Key Challenges

Deliver the right information to the right person at the
right time in the right format




Present only high quality search results




Deliver the results in constrained operation conditions:



Reduced bandwidth



Mobile/text interface



Crisis situation (Human Factors)




Reconcile conflicting reports, inaccuracies, false
information




Integrate data from tens of thousands of real
-
time
sources, in both text and non
-
text formats, which vary in
metadata richness



Government:
Miami
-
Dade County Emergency
Management &
Business Recovery Committee
,
Palm Beach County Emergency Management &
Private
-
Public Partnership,

Broward County
Emergency Management, Monroe
County
Emergency Management


Financial:
FloridaFIRST


Professional Assn:
Greater Miami Chamber of
Commence


Retailers, Logistics, Manufacturing, Services:
IBM,
Office Depot, Wal
-
Mart, Ryder, Greyhound,
Beckman Coulter, NCCI, and many others…


Healthcare
:

South Florida Public Health Institute,
Quantum Group




American Airlines

Aon Corporation

AT&T

Bank of America

Baptist Health South Florida

Becker &
Poliakoff


Carnival Cruise Lines

Catepillar


ClearChannel

Communications

Crowley

Exeter Architectural Products

Florida International University

Florida Public Health Institute

Florida Power and Light

Greyhound Lines

IBM

Home Depot

Hurricane Protection Industries

Insurance Information Institute

City Bank

Beckman Coulter

Miami River Marine Group

*Greater Miami Chamber of
Commerce

Macy’s Florida

Miami Christian School

Miami Dade College

Miami International Airport

Office Depot

Publix Supermarkets

Ryder System

T
-
Mobile

Target Stores

Terremark

Trane

United States Postal Service

UPS

URS Corporation

Visa

Wal
-
Mart

Wachovia Bank

WFOR
-
CBS
-
4

Winn
-
Dixie

Greater Miami Convention & Visitor
Bureau

FloridaFIRST


ABN AMRO

American National Bank

BAC Florida Bank

Banco

Popular

BankFIRST

BankUnited

City National Bank of Florida

CommerceBank
, N.A.

Executive National Bank

Fidelity Federal Bank & Trust

Intercredit

Bank, N.A.

International Bank of Miami,
N.A.

Mellon

Mesirow

Financial

Northern Trust, N.A.

Ocean Bank

Regent Bank

Sun American Bank

Third Federal Savings

Topical Financial Credit Union

Total Bank

Transatlantic Bank

Union Credit Bank

U.S. Century Bank


In disaster situation:

Massive news/reports are generated by different
sources each second.


Disaster related information:

Out of alignment, heterogeneous formats from
multiple channels.


Understand situations of neighborhood:

Geographically or commercially related, dynamic or
static relations.


How can the system quickly capture the status
report information?


Information Extraction


How can the system effectively understand the
situation from a large collection of reports?


Multi
-
document Summarization


How can we automatically capture users’ interests
and effectively deliver the relevant information to
the users?


Dynamic dashboard


How can we take advantage of the community
information for disaster recovery?


Community generation



Effective and interactive information summarization
methods to help users understand large collection
of reports
?


Hierarchical Summarization


Intelligent information delivery techniques to help
users quickly identify the information they need
?


Dynamic query form


Dynamic community generation techniques for
reports recommendation and user group
organization
?


User recommendation


Solutions


Information extraction


Content recommendation


Reports summarization


Community generation


Data Sources

Data Repository



Emergency Operati onal Center

B2B Communi ty

NGOs

Medi a Outl ets

FEMA

Government Servi ce

Reports/Messages/Servi ces/Orders

Indexed Report Fi l e

Recommendati on
Engi ne

Spati al
Cl usteri ng

Database Records

Formatted data input

Informati on extracti on

Unformatted data input






Dynami c
Communi ti es

Dashboard

Documents

summari zati on

Reports Summary

BCiN

-

Business Continuity Information Network

Dashboard

Summarized reports

Reports submit/ File upload


Massive sources at granularities, redundancy


Different varieties: structured and
unstructured

.doc

.pdf

.txt

Email

Disaster
Information
Database

Public Service

-----------------------------------------------------

Address



Open Time

Close Time

Road

-----------------------------------------------



Company

----------------------------------

Address

Asset



Status



Information
extraction


What was/is/will be
the status of
Facilities/Services/…
at the time of…?


Unstructured and
multi
-
source inputs.


Triple <Entity, Time,
Status>.

Time
:

October

21
,

2005

12
:
30

p
.
m
.

Miami
-
Dade

Emergency

Operations

Center

is

currently

activated

at

a

level

II

and

officials

and

emergency

managers

are

carefully

monitoring

Hurricane

Wilma
.


Residents

are

urged

to

finalize

their

personal

hurricane

preparations
.

On

Monday,

October

24
,

Miami
-
Dade

County

offices,

public

schools,

and

courts

will

be

closed
.

Currently,

transit

bus

and

rail

service

continues,

including

Metrobus
,

Metrorail

and

Metromover
.

Miami

International

Airport

is

open
.

However,

if

you

have

travel

plans

please

check

with

your

airline

for

flight

information
.

Tomorrow

afternoon,

the

American

Red

Cross

will

open

hurricane

evacuation

centers

for

residents

who

do

not

feel

safe

in

their

homes

or

live

in

low
-
lying

areas
.



Entity Extraction


Sentence segmentation


POS tagging


Conditional random field



Miami
-
Dade

Emergency

Operations

Center

is

currently

activated

at

a

level

II

and

officials

and

emergency

managers

are

carefully

monitoring

Hurricane

Wilma
.


Residents

are

urged

to

finalize

their

personal

hurricane

preparations
.

On

<T>Monday,

October

24
</T>
,

<E>Miami
-
Dade

County

offices</E>
,

<E>public

schools</E>
,

and

<E>courts</E>

will

be

closed
.

<T>Currently</T>
,

<E>transit

bus</E>

and

<E>rail

service</E>

continues,

including

<E>
Metrobus
</E>
,

<E>Metrorail</E>

and

<E>
Metromover
</E>
.

<E>Miami

International

Airport</E>

is

open
.

However,

if

you

have

travel

plans

please

check

with

your

airline

for

flight

information
.

<T>Tomorrow

afternoon</T>
,

the

American

Red

Cross

will

open

<E>hurricane

evacuation

centers</E>

for

residents

who

do

not

feel

safe

in

their

homes

or

live

in

low
-
lying

areas
.



Relation Extraction


(Entity, Time)
categorization


{ No relation, open,
close, unclear }

Service

Time

Status

Miami
-
Dade

County

offices

October 24, 2005

close

public

schools

October 24, 2005

close

courts

October 24, 2005

close

transit

bus

October 22
, 2005
6:30 p.m.

open

Rail

service

October 22, 2005
6:30 p.m.

open



Miami

International

Airport

October 22, 2005
6:30 p.m.

open

hurricane

evacuation

centers

October 23, 2005
afternoon

open


Content recommendation


Dynamic dashboard


Personalized


Snapshot


Issues to handle


Redundancy


Rank on relevance

Dashboard

EOC
Reports

Company
Reports

News

Company
Messaging

Situation





Threat



Event









Company





The data sources for different dashboard


Content Recommendation Engine


Content ranking


Modeling user preference


I(u): set of information submitted by user


J(u): set of information viewed by user


Scoring the importance



*t(u)
-

tf
-
idf

of document submitted by u.

*pr(d)


the priority of d.


Reports summarization


Multiple reports


Unstructured/Structured information


Sentences: raw, entity/relation identified


Time sensitive


Documents: within 48 hrs. and beyond 48 hrs.


Reports summarization


Summarize recent unstructured reports




Summarize recent structured reports




Most recent sentence associated with entity
information will be picked.


Query:

Airport


151 documents


October 19, 2005

4:30 p.m.
-

November 4,
2005 11:30 a.m.


mi慭i

int敲national ai牰r牴 is
op敮.


homest敡e gen敲al ai牰r牴 is
op敮 from sun物s攠to suns整.


homestead airport remains
closed.


tami慭i

an搠
o灡

locka

airports are open and
operational.


tami慭i

ai牰r牴 is op敮 an搠
operational.


Community


Spatial information is associated with every entity


Geographical region


Similar damage situation


Similar recovery status/interests


Dynamic community


Business relation


Status: open/close


Region divisions


Constraints


Obstacles:


Road forbidden


Curfew area


Facilitators


Open expressway



ADSB API

Hierarchical
Summarization

User Recommendation


Server and Repository

Dynamic Query Form


ADSB API


REST framework


Interact with multiple
information domains


Fast prototype and create
functions


Improve end
-
user
programmability


Easily released and embedded to
3
rd

party clients

ADSB


All
-
hazardous Disaster Information Network


Affinity Propagation


Input: sentence similarity graph defined as



G<V, E>: v
∊ V

is a vertex representing a
sentence, e
∊ E
is an edge.



s(
i,k
): similarity between two distinct vertices


Two types of messages


Responsibility r(
i,k
): how well k is chosen to be
exemplar of
i
.


Availability a(
i,k
): how well that
i

choose k to be its
exemplar.


Objective function


argmax
k
{a(
i,k
) + r(
i,k
)}


Hierarchical Multi
-
document Summarization


Sentence preference


Language model score
L:


short sentence with more frequent words has higher
score.


LexPageRank

score
P:


page rank of sentences on graph


Freshness score
F:


latest information has higher score

Preference score
s(
i,i
)

=
L
i
+ P
i

+ F
i


Sentence similarity


s(
i,j
) =
logP
(
i|j
), means how likely sentence i can be
summarized by sentence j.


Two types of query form
components:


Attribute display


Different attribute sets


Different annotations


Query conditions

Precision
E
(F) =

d
∊D

p
u
(d)
p
u
(A
F
)
p
u
(
σ
F
|d
)/∑
d
∊D

p
u
(
σ
F
|d
),

Recall
E
(F) =

d
∊D

p
u
(d)
p
u
(A
F
)
p
u
(
σ
F
|d
)/∑
d
∊D

p
u
(d)
p
u
(A),

FScore
E
(F) =
((1+
β
2
)⋅
Precision
E
(F)

Recall
E
(F)
)/(
β
2

Precision
E
(F)

Recall
E
(F)
).

F
*

=
argmax
F
Fscore
(F)


Problem Formulation

Query Form F=(
A
F
,

σ
F
), where
A
F
is a set of displaying
attributes,
σ
F

is a set of query conditions.




To evaluate the overall goodness of a query form:

Report Graph

Attribute Graph

p
u
(d)=

d’
∊D
u,f

p
u
(
d|d
’)
p
u
(d’)

p
u
(A
F
i+1
)=
p
u
(
A|
A
F
i
)
p
u
(
A
F
i
)


Random Walk Model

FScore
(F
i+1
):
p
u
(d),
p
u
(A
F
i+1
) and
p
u
(
σ
F
i+1
|d)



Recommendation tasks in ADSB


To form meaningful user groups


People have similar reading preferences


People have common sharing behaviors


To share pieces of information with related people


People are possibly interested in certain information


People are interested in others’ interests


Three important factors


Time


Direction


Textual content

U

1

2

4

3

Transaction Hyper
-
graph


User profile



User preference



Group contribution


Recommend report to users




Recommend users to communities

profile(u)=
α
⋅ ∑
d
∊S
(u)
tf
-
idf
(d)
t
+
β

⋅ ∑
d
∊R
(u)
tf
-
idf
(d)
t

preference(
u,d
)=
cos
(profile(u) ,
tf
-
idf
(d)
t
)

gc
(
d,g
)=
α

⋅ ∑
i
=O(
u,g
)
s(
i,d
)
t
+
β

⋅ ∑
i
=I(
u,g
)
s(
i,d
)
t

,where, s(
i,d
)=

u
∊i

preference(
u,d
).

gc
(
c,g
)=
α

⋅ ∑
i
=O(
u,g
)
s(
i,c
)
t
+
β
⋅ ∑
i
=I(
u,g
)
s(
i,c
)
t

,where, s(
i,c
)=

u
∊i

cos
(profile(u), profile(c)).

Input:
u
, the user;
d
, the report, and

, the seeds

Output:

, recommended user list

1.
𝓖



GetTransactionalGroups
(u)

2.






3. for each group
g


𝓖

4. for each user
c

∈ g,
c




5. if
c




6.


[
c
]


0

7.

[
c
]



[
c
] +
GroupScore
(
c,

, g, d
)


or

[
c
]



[
c
] +
CommunityScore
(
c,

, g
)


GroupScore
(
c,

, g, d
) =
gc
(
d,g
), if

⋂g≠⌀; 0 otherwise.


CommunityScore
(
c,

, g
) =
gc
(
c,g
), if

⋂g≠⌀; 0 otherwise.




Suggesting user routine


Education and
Diversity

Graduated:




Jesus
Bello BS/CS (Hispanic)




Michael
Calleiro

BC/CS (Hispanic)




Allison
Lanager

BS/CS (Female)




Seychelles
Martinez

BS/CS
(
Hispanic/Female)


Awarded Top Undergraduate of School



Priyanka

Bansal
, MS/CS (Female)



Hardik

Dave, MS/CS



Yi
Zhang, Ph.D. Student



Jason Allen, BS/CS (Undergraduate)



Gisselle

Ginarte

, BS/CS (Hispanic/Female)




Alina

Gayazova
,
BS/CS

(Female)


Supported:



Li
Zheng
, PhD
Student



Chao
Shen
, Ph.D.
Student


Liang
Tang, Ph.D.
Student


Amanda Crick (
Female,Black
)


Jingxuan

Li, PhD Student


Hsin
-
Yu Ha, Ph.D.
Student


Jesse
Domack
,
BS/CS (Undergraduate, 2012 GMCC finalist)


Jason Clary, BS/CS (Undergraduate)

Acknowledged:



Enzo

Alvarez, BS/CS (Hispanic)



Mark
Oleson
, BS/CS (Undergraduate)


















L.
Zheng
, C.
Shen
, L. Tang, T. Li, S. Luis, S.
-
C. Chen. Applying Data Mining
Techniques to Address Disaster Information Management Challenges on
Mobile Devices. , ACM KDD, pp. 283
-
291, 2011.



L.
Zheng
, T. Li. Semi
-
supervised Hierarchical Clustering. In Proceedings of
2011 IEEE International Conference on Data Mining.



Liang Tang, Tao Li,
Florian

Pinel
, Larisa
Shwartz
,
Genady

Grabarnik
.
Optimizing System Monitoring Configurations for Non
-
Actionable Alerts. In
Proceedings of IEEE/IFIP Network Operations and Management Symposium
(IEEE NOMS), 2012.



Chao
Shen

and Tao Li. Learning to Rank for Query
-
focused Multi
-
document
Summarization. In Proceedings of 2011 IEEE International Conference on
Data Mining (ICDM 2011).




Fausto

Fleites
,
Kasturi

Chatterjee
, and
Shu
-
Ching

Chen, "AH+
-
Tree: An
Efficient Multimedia Indexing Structure for Similarity Queries," IEEE
International Symposium on Multimedia (ISM2011), Dana Point, California
USA, pp. 69
-
76, December 5
-
7, 2011.




Dianting

Liu, Mei
-
Ling
Shyu
,
Qiusha

Zhu, and
Shu
-
Ching

Chen, "Moving
Object Detection under Object Occlusion Situations in Video Sequences,"
IEEE International Symposium on Multimedia (ISM2011), Dana Point,
California USA, pp. 271
-
278, December 5
-
7, 2011.



Steven Luis,
Fausto

C.
Fleites
,
Yimin

Yang,
Hsin
-
Yu Ha, and
Shu
-
Ching

Chen, "A Visual Analytics Multimedia Mobile System for Emergency
Response," IEEE International Symposium on Multimedia (ISM2011),
Dana Point, California USA, pp. 337
-
338, December 5
-
7, 2011. (Demo
paper)





Lei Li, Li
Zheng
, and Tao Li. LOGO: A Long
-
Short User Interest Integration in
Personalized News Recommendation. In Proceedings of the 2011 ACM
Conference on Recommender Systems (
RecSys

2011), pages 317
-
320, 2011.



Chao
Shen
, Tao Li, and Chris Ding. Integrating Clustering and Multi
-
Document Summarization by Bi
-
mixture Probabilistic Latent Semantic
Analysis (PLSA) with Sentence Bases. In Proceedings of the Twenty
-
Fifth
Conference on Artificial Intelligence (AAAI
-
11), 2011.



Chao
Shen

and Tao Li. A Non
-
negative Matrix Factorization Based Approach
for Active Dual Supervision from Document and Word Labels. In Proceedings
of Conference on Empirical Methods in Natural Language Processing
(EMNLP 2011), 2011.




Liang Tang, Tao Li, and Charles
Perng
.
LogSig
: Generating System Events
from Raw Textual Logs. In Proceedings of the 20th ACM Conference on
Information and Knowledge Management (CIKM 2011).



V.
Hristidis
, S
-
C. Chen, T. Li, S. Luis, Y. Deng. Survey of Data Management
and Analysis in Disaster Situations. Elsevier Journal of Systems and Software
(JSS), Volume 83, Issue 10, pp. 1701
-
1714, October 2010



L.
Zheng
, C.
Shen
, L. Tang, T. Li, S. Luis, S.
-
C. Chen, V.
Hristidis
. Data Mining
Techniques to Address Critical Information Exchange Needs in Disaster
Affected Public
-
Private Networks , ACM KDD, pp. 125
-
134, 2010.



L. Li, D. Wang, C.
Shen
, T. Li. Ontology
-
enriched Multi
-
Document
Summarization in Disaster Management. SIGIR 2010.




V.
Hristidis
, E. Ruiz. CADS: A Collaborative Adaptive Data Sharing Platform.
VLDB Workshop on Personalized Access, Profile Management, and Context
Awareness in Databases (
PersDB

2009).



K.
Saleem
, S. Luis, Y. Deng, S
-
C. Chen, V.
Hristidis
, T. Li. Towards a Business
Continuity Information Network for Rapid Disaster Recovery. 9th Annual
International Conference on Digital Government Research 2008




D. Wang, L.
Zheng
, T. Li, Y. Deng: Evolutionary Document Summarization for
Disaster Management. SIGIR 2009.






L. Tang and T. Li.
LogTree
: A Framework for Generating System Events
from Raw Textual Logs. In Proceedings of 2010 IEEE International
Conference on Data Mining.



L.
Zheng
, T. Li, and C. Ding. Hierarchical Ensemble Clustering. In
Proceedings of 2010 IEEE International Conference on Data Mining.



C.
Shen
, D. Wang, and T. Li. Topic Aspect Analysis
forMulti
-
Document
Summarization. In Proceedings of the 19th ACM International Conference on
Information and Knowledge Management (CIKM 2010).



D. Wang, T. Li, S. Zhu, and Y. Gong.
iHelp
: An Intelligent Online Helpdesk
System. IEEE Transactions on Systems, Man, and Cybernetics Part B,
accepted, 2010.



L. Li, D. Wang, C.
Shen
, and T. Li. Ontology
-
Enriched Multi
-
document
Summarization in Disaster Management. ACM SIGIR , Pages 819
-
820, 2010.



M.
-
L.
Shyu
, C. Chen and S.
-
C. Chen, "Multi
-
Class
Classificatio

via Sub space
Modeling," International Journal of Semantic Computing, in press.



DHS, “A Research and Educational Framework to Advance Disaster Information
Management in Computer Science PhD Programs”



Eugenio
Pino

and Family Global Entrepreneurship Center, “All
-
Hazard Disaster
Situation Browser (ADSB) on Mobile Devices”



DHS, "Information Delivery and Knowledge Discovery for Hurricane Disaster
Management”



NSF CREST, "CREST: Center for Innovative Information Systems Engineering”



NSF, “CAREER
: A Collaborative Adaptive Data Sharing
Platform”



Purdue University/DHS VACCINE
CoE
,

“A
Data Mining Framework for Enhancing
Emergency Response Situation Reports with Multi
-
Agency Multi
-
Party Multimedia
Data”



IBM Faculty Shared University Research Award

45


Shu
-
Ching

Chen (
chens@cis.fiu.edu
)

Steve Luis (
luiss@cis.fiu.edu
)