Public expects a prompt response from online services, including emergency response organizations to requests for help posted on social media. However, the information overload experienced by these organizations, coupled with their limited human resources, challenges them to timely identifying and prioritizing such requests. We present a novel model to formally characterize social media requests and then, develop a Learning-to-Rank system using this model.
Paper: Purohit, H., Castillo, C., Imran, M., and Pandey, R. (2018). Social-EOC: Serviceability Model To Rank Social Media Requests for Emergency Operation Centers. ASONAM 2018.
Automatically Rank Social Media Requests for Emergency Services using Serviceability Model - ASONAM18
1. Social-EOC: Serviceability Model To Rank Social Media
Requests for Emergency Operation Centers
Hemant Purohit, Rahul Pandey
Humanitarian, Semantics &
Informatics Lab (Human_Info_Lab)
The 2018 IEEE/ACM International Conference on
Advances in Social Networks Analysis and Mining (ASONAM-2018)
Barcelona, Spain
Aug 29, 2018
Carlos Castillo
Web Science & Social
Computing Research Group
Muhammad Imran
Crisis Computing & Social
Computing Group
@hemant_pt @chaToX @mimran15 @gizmowiki
2. Serviceability Model for Social Media Services, ASONAM-18
Outline
¨ Introduction
n Social media & web streams during disasters
n Problem: filter and rank requests for help
¨ Background and Related Work
n Services in emergency management
n Social computing for emergency management
n Mining request / help-seeking intent
¨ Social-EOC: Serviceability Model
n Qualitative and quantitative formulation
n Social-EOC based Learning-to-Rank system
¨ Experiments
n Conversational data collection
n Learning-to-Rank evaluation
¨ Results
¨ Discussion and Future Work
2
3. Serviceability Model for Social Media Services, ASONAM-18
Social Media & Web Use in Disasters
3
When traditional call systems are exhausted ..
Source: https://www.usatoday.com/story/news/nation-now/2017/08/27/desperate-help-flood-victims-houston-turn-twitter-rescue/606035001/
4. Serviceability Model for Social Media Services, ASONAM-18
Mining Social & Web Data: Innovation
Opportunity for Emergency Services
4
But ..
WHAT & HOW
to filter?
5. Serviceability Model for Social Media Services, ASONAM-18
Mining Social & Web Data: Relevancy issue
5
6. Serviceability Model for Social Media Services, ASONAM-18
Mining Social & Web Data: Ranking issue
(Anonymized) Message Serviceable Degree
@_USER_ I am 9 ft above current water levels,
why am I told to evacuate Grand Lakes now?
Please advise.
serviceable
@_USER_ If there has been no rain since
yesterday, why is water not draining?
Serviceable but lacks details
@_USER_ Thank God you are working on this.
Let us chat when things settle down
not serviceable
¨ Illustration from Hurricane Harvey 2017
6
7. Serviceability Model for Social Media Services, ASONAM-18
Problem
7
¨ Can we filter/classify and prioritize/rank serviceable
social media requests for emergency services?
¨ Scope:
¤ Study requests directly sent to the accounts of services
¤ Analyze the requests during a disaster event
8. Serviceability Model for Social Media Services, ASONAM-18
Mining Social & Web Data: Challenges
¨ Variable Actionability
¤ Subjective relevance of messages, often task-
dependent
¨ Insufficient Samples
¤ Limited context representativeness of request samples
to directly learn ranking
8
9. Serviceability Model for Social Media Services, ASONAM-18
Solution: Our Contributions
9
¨ Novel generalizable model (Social-EOC) to define serviceability
characteristics of a service request message on social media
¨ Learning-to-rank system based on the Social-EOC model
¤ Inference of the serviceability characteristics
¤ Classification and ranking of serviceable requests
¨ Evaluation of the Social-EOC model with the datasets from six
crisis events
10. Serviceability Model for Social Media Services, ASONAM-18
Outline
¨ Introduction
n Social media & web streams during disasters
n Problem: filter and rank requests for help
¨ Background and Related Work
n Services in emergency management
n Social computing for emergency management
n Mining request/ help-seeking intent
¨ Social-EOC: Serviceability Model
n Qualitative and quantitative formulation
n Social-EOC based Learning-to-Rank system
¨ Experiments
n Conversational data collection
n Learning-to-Rank evaluation
¨ Results
¨ Discussion and Future Work
10
11. Serviceability Model for Social Media Services, ASONAM-18
Services in Emergency Management (EM)
¨ Incident-Command System (ICS) model for the response coordination
11
Public Info. Officer (PIO)
provides the information
services [FEMA; Hughes & Palen,
JHSEM 2012]
FEMA: https://training.fema.gov/ programs/pio/ Source: https://en.wikipedia.org/wiki/Incident_Command_System
12. Serviceability Model for Social Media Services, ASONAM-18
Social Computing for EM: Crisis Informatics
12
¨ Extensive literature on social media during disasters [Castillo,
Cambridge Press 2016; Imran et al., CSUR 2015]
¤ Addressed a variety of problems [c.f. Tutorial: Castillo, Diaz & Purohit, SDM 2014]
n Data collection and filtering[e.g., Olteanu et al., ICWSM 2014]
n Modeling actionable help behavior [e.g., Purohit et al., SocialCom 2015]
n Summarization [e.g., Rudra et al., HyperText 2016]
n Information diffusion [e.g., Starbird & Palen, ISCRAM 2010]
n Information and source credibility [e.g., Castillo et al., WWW 2011]
n Visual analytics [e.g., Kumar et al., ICWSM 2011]
n ..
¨ Open problem: finding formal characteristics of relevant
social media requests that must be prioritized
13. Serviceability Model for Social Media Services, ASONAM-18
Mining Request/Seeking Intent
13
REQUEST MEDIUM TYPE FOCUS
Emails
Finding and ranking messages to reply with
different interpersonal communication
behavior for information seeking
[Yang et al., SIGIR 2017; Lampert et al., HLT 2010]
Q&A Forums
Generic seeking behavior for all types of
users and often not targeted towards
answers from a specific organization/group
[Mai, Emerald 2016; Vasilescu et al., CSCW 2014]
Social Media Chats
Finding explicit or implicit requests for help
during disaster relief and users who could
answer queries [Purohit et al., First Monday 2013;
He et al., WebSci 2017; Ranganath et al., TKDE 2017;
Sachdeva & Kumaraguru, CSCW 2017]
14. Serviceability Model for Social Media Services, ASONAM-18
Outline
¨ Introduction
n Social media & web streams during disasters
n Problem: filter and rank requests for help
¨ Background and Related Work
n Services in emergency management
n Social computing for emergency management
n Mining requests / help-seeking intent
¨ Social-EOC: Serviceability Model
n Qualitative and quantitative formulation
n Social-EOC based Learning-to-Rank system
¨ Experiments
n Conversational data collection
n Learning-to-Rank evaluation
¨ Results
¨ Discussion and Future Work
14
15. Serviceability Model for Social Media Services, ASONAM-18
Social-EOC Serviceability Model:
Qualitative Domain Knowledge based on FEMA PIO Guide
15
Explicit
Request
E(m)
Answerable
Query
A(m)
Sufficiently
Detailed
D(m)
Correctly
Addressed
C(m)
Serviceability(m) = f ( E(m), A(m), D(m), C(m) )
Explicitly asks for a
resource or service
Explicitly asks a question
that can be answered
Sent to organization or
person who could have
resources or provide the
service, an alarm, or
could answer questions
Specifying contextual
information: time (when),
location (where),
quantity (how much),
resource (which)
16. Serviceability Model for Social Media Services, ASONAM-18
Serviceability Model: Quantifying Characteristics
(Anonymized) Message Explicit Answer-
able
Addressed Detailed
@account1 please, governor, post a phone # for
specific info in our local areas
4.3 4.3 3.3 3.7
@account2 is thr parking at McMahon for volunteer? 4.0 5.0 5.0 5.0
@account3 how can I help 1.3 4.3 4.3 1.0
@account4 Plz pray for these families 1.7 1.0 1.0 1.0
@account5 been working in #LAFlood shelter, we
actively monitor SM for feedback
1.0 1.0 2.0 2.0
“@account7 No matter where in the world ur
followers live, you can donate from link Plz RT
1.0 1.0 1.0 1.0
¨ E(m), A(m), C(m), D(m) : Likert Scale Functions [score:1-5]
16
Illustration Table: Average scores of Likert ratings by crowd annotators
17. Serviceability Model for Social Media Services, ASONAM-18
Learning-to-Rank System Design: Using Social-
EOC Model
17
18. Serviceability Model for Social Media Services, ASONAM-18
Learning-to-Rank System Design: Steps
18
1. Collecting conversation chains of seed messages
n A seed message mentions a targeted account of official services (e.g.,
@fboem for ‘Fort Bend County OEM’) identified using gov./news reports
2. Rating serviceability characteristics of a request
n Also additional category ‘Other’: advertisements, jokes, ..
n Annotation by crowdsourcing workers
3. Creating gold standard of serviceable requests
n Binary classes: Serviceable (relevant), Not-Serviceable (Not relevant)
n Annotation by Emergency Management professionals
4. Learning to classify and rank serviceable requests
n SVM-Rank
n Relevance grade levels: binary
19. Serviceability Model for Social Media Services, ASONAM-18
Outline
¨ Introduction
n Social media & web streams during disasters
n Problem: filter and rank requests for help
¨ Related Work
n Services in emergency management
n Social computing for emergency management
n Mining requests / help-seeking intent
¨ Social-EOC: Serviceability Model
n Qualitative and quantitative formulation
n Social-EOC based Learning-to-Rank system
¨ Experiments and Results
n Conversational data collection
n Learning-to-Rank evaluation
n Results
¨ Discussion and Future Work
19
20. Serviceability Model for Social Media Services, ASONAM-18
Experiments
20
¨ Data
¤ Collected Twitter messages related to 6 crisis events:
n Big: Hurricane Harvey 2017, Nepal Earthquake 2015, Alberta Floods 2013
n Small: Louisiana Floods 2016, Oklahoma Tornado 2013, Hurricane Sandy 2012
¤ Crawl Twitter conversation chains for seed messages
21. Serviceability Model for Social Media Services, ASONAM-18
Experiments
21
¨ Feature (f.) extraction for Learning-to-Rank method
¤ Generic f.: number of words, hashtags, mentions, and URLs in a tweet
¤ Text f.: tf-idf with Bag-of-Words representation
¤ Manual serviceable f.: average annotation rating between 1 to 5
for each serviceable characteristic
¤ Inferred serviceable f.: 0 or 1
n Binary classifier for each characteristic (rating 1-2: negative, 3-5: positive
class}
¨ Evaluation metric for top-K ranked results
¤ Normalized Discounted Cumulative Gain (NDCG)
¤ NDCG @10
22. Serviceability Model for Social Media Services, ASONAM-18
Experiments: Evaluation Schemes
22
LTR models based on varied features (f.):
¨ [T]: Text f. and generic f. (baseline)
¨ [T+I]: T and inferred serviceable f. from the model built on the
same event data
¤ [T + I_all]: T and inferred serviceable f. from the model built
on the all events data
¤ [T + I_cross]: T and inferred serviceable f. from the model
built on the cross events data (all except the current)
¤ [T_cross + I_cross]: both T and inferred serviceable f. from the
model built on the cross events data (all except the current)
¨ [T+M]: T + manual serviceable f.
23. Serviceability Model for Social Media Services, ASONAM-18
Results
1. Models with inferred
serviceable features
are generally better
than the baseline.
2. Performance varies in
cases of small datasets.
n less training examples
n imbalanced
23
24. Serviceability Model for Social Media Services, ASONAM-18
Results
3. Cross-event models
perform well.
n especially for smaller
datasets
4. Serviceability
characteristics based
features are among the
best discriminators.
n among the top-5 features,
identified using χ2 test
24
25. Serviceability Model for Social Media Services, ASONAM-18
Discussion: Examples of Resulting Ranked Requests
25
Ranked Messages by T (text)+I (Inferred) Modeling Scheme
TOP-2
[Sandy]
- @_USER_ please, governor, post a website or phone# where we can get
specific info for our local areas
- @_USER_ Queens trains aren’t being addressed at all. When can v expect any
service updates for the NQR trains?
BOTTOM-2
[Sandy]
- @_USER_ Romney not going2like that gov christie is being nice about Obama’s
leadership
- @_USER_ HILARIOUS! That’s much needed laughter, I am sure.
TOP-2
[Alberta]
- @_USER_ can you tell me if sanitary pumps are running yet in elbow park?
#yycflood
- @_USER_ plz text with what you need & address. Lots of volunteers in mission
BOTTOM-2
[Alberta]
- @_USER_ thank u calgary police
- @_USER_ Tx for ur time!!
26. Serviceability Model for Social Media Services, ASONAM-18
Discussion: Future Application for Prioritization of
Streaming Messages
26
Serviceability
Source: https://blog.bufferapp.com/twitter-timeline-algorithm
BEYOND TIME & CREDIBILITY,
RANK BY
27. Serviceability Model for Social Media Services, ASONAM-18
Discussion: Lessons, Limitations and Future Work
27
¨ Lessons learned:
n Serviceability characteristics capture the notion of relevance and
serviceability for social media requests to organizational services.
n Experimented for binary relevance (serviceability) grades, but the method is
extensible for multiple grade levels.
¨ Limitations & opportunities for future work
n Study non-English language request messages
n Explore multiple as opposed to single platform based datasets
n Include indirectly addressed request messages (i.e. not starting with @user)
n Generate optimal request alerts to respond
n Our new work: Purohit, Castillo, Imran, Pandey (2018, to appear). Ranking of Social
Media Alerts with Workload Bounds in Emergency Operation Centers. Web Intelligence.
28. Serviceability Model for Social Media Services, ASONAM-18
Conclusion
¨ Demonstrated that social media requests have some common
core characteristics for helping rank the serviceable requests
for organizational services.
¨ Presented a novel qualitative-quantitative model for
serviceable request characteristics for emergencies and
demonstrated its use by Learning-to-Rank (LTR) methodology.
¨ Evaluated LTR systems based on inclusion of serviceability
characteristics features across six disaster events and noted a
superior performance (gain up to 25% in nDCG@10 and nDCG@5).
28
29. Serviceability Model for Social Media Services, ASONAM-18
PAPER: http://ist.gmu.edu/~hpurohit/informatics-
lab/papers/serviceability_ranking_disasters_ASONAM18_final.pdf
CONTACT: hpurohit@gmu.edu
Acknowledgement:
image sources, collaborators (especially Profs. Amit Sheth, Valerie Shalin, & TK Prasad at Kno.e.sis Center;
U.S. DHS Science & Technology SMWGESDM Researcher-Practitioner Subgroup), Human_Info_lab Alumni
(Yogen, Sharan) as well as sponsors:
Questions?
29
Grants:
IIS #1657379,
IIS #1815459
La Caixa project:
LCF/PR/PR16/11110009