Transcript OMG, from here, I can see the flames!” A use

ACM LBSN '09, Seattle, WA, USA, November 3rd 2009
1
“OMG, from here, I can see the flames!”
A use case of mining Location Based Social
Networks to acquire spatio-temporal data on
forest fires
by Bertrand De Longueville*, Robin S. Smith* and Gianluca Luraschi*
* Joint Research Centre (JRC)
IES - Institute for Environment and Sustainability
Spatial Data Infrastructures Unit
http://sdi.jrc.ec.europa.eu/
“OMG, from here, I can see the flames!” A use case of mining Location
Based Social Networks to acquire spatio-temporal data on forest fires.
ACM LBSN '09, Seattle, WA, USA, November 3rd 2009
Need
Forest fires: know you enemy

Forest Fires are a major natural hazard in Europe:
•
•
•
35000 forest fires reported annually
5000 km2 burnt annually (about 2000 square-miles)
Causes multiple forms of societal, environmental and economic
damage
Fire prevention and -fighting relies on
accurate and timely data for:

•
•
•
•
Risk assessment
Early warning
Crisis management
Damage assessment
http://www.eea.europa.eu/highlights/forest-fires-in-southern-europe-destroy-much-more-than-trees
http://effis.jrc.ec.europa.eu
2
“OMG, from here, I can see the flames!” A use case of mining Location
Based Social Networks to acquire spatio-temporal data on forest fires.
ACM LBSN '09, Seattle, WA, USA, November 3rd 2009
Opportunity
Technological innovations are enabling new sources of information
Web 2.0 is introducing a new
communication paradigm, where users can
be also information providers

GPS- and web- enabled mobile devices
are becoming more mainstream


“citizens as sensors”, provides
Volunteered Geographic Information (VGI)
LBSN can be seen as a framework with a
huge potential to create such VGI

O'Reilly, T. (2005). What Is Web 2.0 Design Patterns and Business Models for the Next Generation of Software.
http://www.oreillynet.com/pub/a/oreilly/tim/news/2005/09/30/what-is-web-20.html
Michael T. Jones, “Unexpected Change - Transforming the GeoWeb for a New World,” in Keynote speech presented at the
Geoweb 2009, Vancouver, BC, Canada, 31st July 2009
Goodchild, M. F. (2007). Citizens as Voluntary Sensors: Spatial Data Infrastructure in the World of Web 2.0. International
Journal of Spatial Data Infrastructures Research, 2, 24-32
3
“OMG, from here, I can see the flames!” A use case of mining Location
Based Social Networks to acquire spatio-temporal data on forest fires.
ACM LBSN '09, Seattle, WA, USA, November 3rd 2009
Challenge
What credibility for VGI? What value?

VGI is heterogeneous
Traditional quality control
mechanisms do not apply to
VGI

What is the value of VGI
compared to traditional
information channels?

Flanagin, A. J. and Metzger, M.J. “The credibility of volunteered geographic information,” GeoJournal 72 (2008): 137-148.
4
“OMG, from here, I can see the flames!” A use case of mining Location
Based Social Networks to acquire spatio-temporal data on forest fires.
5
ACM LBSN '09, Seattle, WA, USA, November 3rd 2009
Our Goal
To extract quality-controlled spatiotemporal information from Web 2.0
sources that can be used to
complement existing information
workflows on forest fires
© Axelrauc (via Flickr)
“OMG, from here, I can see the flames!” A use case of mining Location
Based Social Networks to acquire spatio-temporal data on forest fires.
ACM LBSN '09, Seattle, WA, USA, November 3rd 2009
6
Use case
mining the Twitter LBSN to extract spatiotemporal information about the Marseille
forest fire (22-23 Jul. 2009)
“OMG, from here, I can see the flames!” A use case of mining Location
Based Social Networks to acquire spatio-temporal data on forest fires.
7
ACM LBSN '09, Seattle, WA, USA, November 3rd 2009
Context: the Marseille fire
Marseille: situated on the Mediterranean coast, the 2nd most
populated city in France (1.6 million inhabitants)

The fire threatened densely populated neighbourhoods, caused
many people to leave their homes and places-of-work and attracted
the attention of the international media (“the fire at Marseille’s
gates”)

during this event
(from 22nd to 23rd July),
about 1000 ha of forests
were burnt and 10 houses
destroyed

©Matthieu Lestrade (via linternaute.com)
“OMG, from here, I can see the flames!” A use case of mining Location
Based Social Networks to acquire spatio-temporal data on forest fires.
ACM LBSN '09, Seattle, WA, USA, November 3rd 2009
8
Material: tweets about the Marseille fire
Twitter API query: tweets containing the word “incendie” (= fire that
causes important damage) published between 2009-07-22 12:00:00
(GMT+2) and 2009-07-23 12:00:00 (GMT+2)


Result: 313 Tweets from 127 individual users
(total API response was 346 Tweets, but 33 irrelevant Tweets were
easily identified and removed from the analysis)

“OMG, from here, I can see the flames!” A use case of mining Location
Based Social Networks to acquire spatio-temporal data on forest fires.
9
ACM LBSN '09, Seattle, WA, USA, November 3rd 2009
Temporal dynamics
Tweets
"NEWS: An important forest fire
is spreading on the hills, east of
Marseille http://***"
"See how a forest fire looks like,
from Marseille's Vieux Port:
http://***"
"Call for witnesses! Marseille:
fire near the Mont Latin, in an
unpopulated zone."
0
Key Events
2009-07-22
12:00:00
2009-07-22
14:00:00
2
0
2009-07-22
16:00:00
4
3
2009-07-22
18:00:00
[The fire crosses the pass of the
Mont Latin and starts
descending towards Marseille]
[The fire starts near a Military
camp]
"To all our friends in Marseille: Keep
your chin up! #incendie"
"NEWS: Marseille fire: more than
300 inhabitants evacuated near
Trois-Ponts http://***"
"OMG! The fire seems out of
control: It's running down the
hills! No need for TV tonight!"
"An important fire started near
the military camp of Carpiagne
http://*** La Provence.com"
"FRANCE: 1200 hectares burnt in a
huge forest fire threatening
Marseille"
2009-07-22
22:00:00
2009-07-23
00:00:00
"There was more than 50
participants on the chat #marseille
#incendie http://****"
9
2009-07-23
02:00:00
[The fire is at 1km from densely
populated areas, 350 persons
from closer, isolated houses are
evacuated]
[The fire is at 1 hour from the
closest populated areas.]
44
41
19 17
13 10 15 17
2009-07-22
20:00:00
"RT @X All the picture of this
night's #incendie in #marseille on
Flickr! http://*** Frightening!"
"@X @Y RT@Z #incendie
#marseille Check out this Blog
Post: http://***(via @W)"
1
1
2009-07-23
04:00:00
3
50
27
17
15
5
2009-07-23
06:00:00
2009-07-23
08:00:00
2009-07-23
10:00:00
2009-07-23
12:00:00
[The fire is under control but
not 100% put out.]
[Hundreds of citizens from
south-east suburbs evacuate
sponaneously, frightened by
flames and smoke.]
(Sources: information provided during and after the fire by AFP and La Provence, and
selected Twitter contents)
“OMG, from here, I can see the flames!” A use case of mining Location
Based Social Networks to acquire spatio-temporal data on forest fires.
10
ACM LBSN '09, Seattle, WA, USA, November 3rd 2009
Spatial dynamics (1/2)
2 sources of GI: user location in profile or place names (or
coordinates) in text
 user location : only 5 users (out of 127) had GPS-enabled devices;
this location information needs to be collected in real time
 place names : 281 Tweets (out of 313) contained a place name (see
next slide)
 Other spatial information : total burnt area

1400
1200
1000
ha
Values
associated with
the keyword
‘hectares’ in
Tweets over
time
800
600
400
200
0
12 :0 0
16 :0 0
2 0 :0 0
0 0 :0 0
t
0 4 :0 0
0 8 :0 0
12 :0 0
“OMG, from here, I can see the flames!” A use case of mining Location
Based Social Networks to acquire spatio-temporal data on forest fires.
ACM LBSN '09, Seattle, WA, USA, November 3rd 2009
Spatial dynamics (2/2)
11
“OMG, from here, I can see the flames!” A use case of mining Location
Based Social Networks to acquire spatio-temporal data on forest fires.
12
ACM LBSN '09, Seattle, WA, USA, November 3rd 2009
Social dynamics: who tweets?

3 actor-types have been defined
•
•
•

Citizens: physical persons acting on their own behalf
Media: well known traditional media (newspapers, TV networks,
radio)
Aggregators: systems that compile existing information into
specific news-feeds and broadcast them to a targeted audience
Classification ‘by hand’, analysing profile page contents
Number of users that published tweets by type
Number of tweets published by user type
25 (20%)
Citizens
97 (31%)
Citizens
Media
Media
Agregators
20 (16%)
173 (55%)
82 (64%)
43 (14%)
Agregators
“OMG, from here, I can see the flames!” A use case of mining Location
Based Social Networks to acquire spatio-temporal data on forest fires.
ACM LBSN '09, Seattle, WA, USA, November 3rd 2009
13
Social dynamics: How do they tweet?

Re-tweets (tweets containing the RT syntax)
•
•

about 1 tweet out of 5 (18.8%) is a ‘re-tweet’
91% of re-tweets have been published by citizens
Hash tags (tweets tagged with #keywords)
•
•
•
•
about 1 tweet out of 5 (22%) contain hash tags
89% of hash-tagged tweets have been published by
citizens
Nearly all tweets with hash tags contained #incendie
and #marseille
No specific hash tag for the event was spontaneously
created (e.g., #obamainaug, #bneflood, #grfires )
“OMG, from here, I can see the flames!” A use case of mining Location
Based Social Networks to acquire spatio-temporal data on forest fires.
14
ACM LBSN '09, Seattle, WA, USA, November 3rd 2009
URLs analysis
Number of unique cited URLs by domain type
Where do tweets point to?
19 (13%)
Forum, Blogs,
Chats
11 ( 7%)
Media
75% of tweets contained a URL
 Those 236 links pointed towards
148 unique pages and towards
62 unique domains
 4 domain types have been defined
News Portal

53 (36%)
Social Media
65 (44%)
•
•
•
•

Forum, blog, chats (e.g., blogspot.com, tinychat.com)
Social Media (e.g., flickr.com, twitpic.com)
Media (e.g., france-info.com, lemonde.fr, tf1.lci.fr)
News Portals (e.g., marseille-news.com, catnat.net)
Classification ‘by hand’, analysing the domain home page
“OMG, from here, I can see the flames!” A use case of mining Location
Based Social Networks to acquire spatio-temporal data on forest fires.
ACM LBSN '09, Seattle, WA, USA, November 3rd 2009
Discussion
15
- are media hyperboles and previous search verified in this case?
H1: Twitter is an extremely fast information dissemination platform to report
exceptional events.
 The Marseille fire was firstly reported by local newspaper’s web site

H2: As an LBSN, Twitter will provide accurate and useful spatiotemporal
information.

 We found interesting spatiotemporal information; but what value?
H3: Users use Twitter to communicate with each other in widely open
conversation; as a result, it is a primary source of information from citizens

 Yes, but in our case it was ‘diluted’ in redundant information from media
H4: Twitter is used as information broadcasting and brokerage platform during
crisis events

 Yes, the Marseille fire provided a typical example of such use of Twitter
Hughes, A.L. and Palen, L. Twitter Adoption and Use in Mass Convergence and Emergency Events. Proceeding of the 6th
International ISCRAM Conference, (2009)
Johnson, S. How Twitter Will Change the Way We Live. Time, 2009-06-05.
Ulrich, C. Twitter, média de l'ère Obama. Le Monde 2 - special Hi-Tech, 2008-11-14
“OMG, from here, I can see the flames!” A use case of mining Location
Based Social Networks to acquire spatio-temporal data on forest fires.
16
ACM LBSN '09, Seattle, WA, USA, November 3rd 2009
Conclusion
LBSN as a real-time source for spatio-temporal event-related information
Tweets may contain valuable spatiotemporal data (and even more in the
future)

The discrimination between primary
and secondary information is uneasy
but crucial

Tweets contain many links that can
be a good starting point for crawling

Source: http://www.iphonesavior.com/
Twitter offers spontaneously a
timeline for event-related information

“OMG, from here, I can see the flames!” A use case of mining Location
Based Social Networks to acquire spatio-temporal data on forest fires.
ACM LBSN '09, Seattle, WA, USA, November 3rd 2009
17
Conclusion
LBSN as a real-time source for spatio-temporal event-related information
Just as we readily accept the
processing of satellite data as an
input to many geospatial
analyses, we should also aim to
better interpret the abundant and
freely available signals provided
by citizen-sensors.
© Ian Hanning/AFP
“OMG, from here, I can see the flames!” A use case of mining Location
Based Social Networks to acquire spatio-temporal data on forest fires.
ACM LBSN '09, Seattle, WA, USA, November 3rd 2009
18
Thank you for you attention
[email protected]
Source: http://geekandpoke.typepad.com/