More than Just Lines on a Map: Best Practices for U.S Bike Routes
User generated data on extreme events
1. 1 Dept. of Systems Engineering & Computer Science, Universidad del Norte, Colombia
2 Dept. of Civil and Environmental Engineering, Politecnico di Milano, Como Campus, Italy
3 Transport and ICT Global Practice, The World Bank
User Generated Data During Extreme
Events
Mayra Zurbarán1, Maria A. Brovelli2, Danilo Ardagna2,
Mattia Manara2, Mark Iliffe3
2. Social media APIs allow access to rich sources of user generated content
This serves for manyfold applications, specially when it is geo-referenced
data, e.g.:
Spotting popular locations
Different activities within a city
Relevant local news
Sentiment analysis
Monitoring extreme events
This was done in Italy using the Twitter Streaming API for the following
events:
Monitor precipitation in the country
Earthquake detection
Script openly available at: https://github.com/mazucci/geocollect
3. The Twitter Search API:
• Size limited response per
request
• Needs to reconnect on every
query
• Restricted requests over
time; limited at 180 queries
per 15 min window
Available at:
https://dev.twitter.com/rest/public/
search
The Twitter Streaming API:
• Needs only one request and the
connection remains
• Low latency access up to an
estimated 1% of all twitter data
• Parametrized queries get higher
percentage of the full response
• Optimized traffic due to less
connection attempts
Available at:
https://dev.twitter.com/streaming/over
view
For Italy around 30,000 geo-referenced tweets are collected daily
through the Streaming API
7. Hashtag Appered First At Number of Tweets
#Italy 03:40 59
#terremoto 03:40 287
#earthquake 03:40 149
#sismo 03:41 75
#Roma 03:41 5
#terremotoRoma 03:41 2
#quake 03:56 3
#Norcia 04:09 5
#terremotoItalia 8:38 7
First reported tweet after the earthquake at aprox. 250 km from the
epicentre:
2 minutes after the official
time of major earthquake
(Mag: 6,0) at 3:36 AM
Popular Hashtags
9. A total of 498 tweets were related to the earthquake on Aug 24th
0
10
20
30
40
50
60
70
80
90
Roma
Perugia
Rieti
Ragusa
Savona
AscoliPiceno
Palermo
Fermo
Ancona
Napoli
Teramo
Chieti
Macerata
Milano
Firenze
Torino
Modena
Salerno
Pescara
Cosenza
Rimini
Genova
Forli'-Cesena
Pistoia
Terni
Viterbo
Venezia
Tweet Count by Provinces (IT)
Count
12. *Values calculated using the Kernell formula by Silverman (1986, p. 76, equation 4.5)
Data: A total of 499
earthquakes reported
from Aug 17th until Aug
24th of 2016.
Source: INGV -
http://www.ingv.it/it/
15. Keywords to include while processing:
Temporale, acquazzone, diluvio, alluvione
16. During extreme events, in our case study of earthquakes in central
Italy:
Users tweet more when they are near the epicentre but on non
critical areas
Population density impacts the content generation: There are
more tweets from big cities
Aggregating user-generated content aids in understanding user’s
perceptions during extreme events e.g. EMSC - http://www.emsc-
csem.org
The precipitations study shows that are more tweets regarding rain
near the Alps Area where it rains more during summer time
according to official reports
For Future Work it is interesting to supervise the behavior during
different seasons