IEEE Consumer Electronics Magazine - November/December 2021 - 94

Special Section on Advances in Smart Cities
Figure 3. Map showing the location of categorized feeds for Rain (left) and AEMETs reflectivity map on
January 21, 2020 (right).
of this information provides a real-time channel
of information about the state and consequences
of the natural disaster.
Figure 1 shows the number of feeds that were
categorized in each source, along with the time
when those feeds were published and their
impact, being categorized using the PPI metric.
This figure shows important information that
really matches the results displayed in Figure 2,
where the two control points are also represented.
Here, the feeling is also introduced
because people can write about a particular
topic in a positive or negative way. However, the
application is specifically aimed at detecting
problems, so we are interested in very negative
opinions (i.e., the darkest area in Figure 1).
Finally, Figure 3 shows two maps. The map on
the left displays the location of the categorized
feeds for the category Rain. They were located
by making use of the toponyms mentioned in the
microtexts. The map on the right is the reflectivity
map generated by the official State Meteorological
Agency (AEMET). Reflectivity images
correspond to the lowest elevation of the radar
scan (0:5) above the horizontal plane), and the
color scale indicates reflectivity ranges in Z decibels.
It should be noted that a correlation exists
between these maps, which were generated for
the same time frame (i.e., from January 21, 2020
at 00:30 to January 21, 2020 at 12:00). Indeed, as
clusters of hazard-related posts correspond to
areas with high decibel levels, we conclude that
there is a direct correlation between people's
opinions and natural events.
94
CONCLUSIONS
This article presented WATERSensing,a
social sensing application for the early detection
of water-related issues through people's
opinions in social networks. We introduced
the software architecture underlying this application,
which is based on three main modules:
1) data collection from different social networks
(crawling module), 2) microtext processing to
semantically analyze the information retrieved
by our system (NLP module), and 3) visual feedback
to the user to look for insights (visualization
module). The NLP module establishes
several problem-identification metrics that are
used to classify tweets based on how likely they
reflect a given problem. Moreover, we designed
an NER module that recognizes several named
entities, such as locations, persons, and organizations.
Location entities were disambiguated
at the visualization stage by pinpointing locations
on a map.
To assess our approach, we analyzed Storm
Gloria, which occurred on the eastern Mediterranean
coast from January 19, 2020 to January 22,
2020. It should be noted that both the alerts
issued by the Spanish government before the
event and the catastrophic events that occurred
had an impact on people's opinions on social
networks. The results conclude that the perception
of the problem in social networks and the
media perfectly correlates with the evolution of
the natural disaster. We can anticipate that the
benefits of our approach would be relevant to
other natural disasters, and the combination
IEEE Consumer Electronics Magazine

IEEE Consumer Electronics Magazine - November/December 2021

Table of Contents for the Digital Edition of IEEE Consumer Electronics Magazine - November/December 2021