IEEE Consumer Electronics Magazine - November/December 2021 - 93

Figure 1. Number of messages in the Rain space in the last 24 h according to a sentiment analysis from
January 19, 2020 (14:00) to January 20, 2020 (14:00).
Figure 2. Influx of messages in the Rain space in
the last 24 h [from January 19, 2020 (14:00) to
January 20, 2020 (14:00)]. Blue area shows Twitters
messages and the Orange one RSS messages.
date, but also by relevance, feeling, PPI, etc.
There is also a search bar where users can
search for particular keywords. To speed up the
search process, the entities recognized with the
NER are displayed in the search bar, so that the
user can select which texts contain a certain
entity. The application also shows an image gallery
that contains the pictures accompanying
the microtexts that were successfully categorized.
Finally, some diagrams are displayed to
show the volume of processed messages in this
topic, the feeling of these messages, and their
localization.
RESULTS AND DISCUSSION
Our tool can process information in real time
and provide several metrics such as the number
of tweets, the PPI for each tweet, and the most relevant
users. The results shown in Figures 1 and 2
are based on the space Lluvia (Rain), with the following
keywords associated to the space: torrente,
chubasco, aguacero, chaparron, diluvio,
torrencial, anegar, inundar, llover a cantaros,
diluviar, diluvial, precipitacion, lluvıa, pluvioso,
lluvioso, llover, precipitar, tromba, and marina
(i.e., rainfall, downpour, cloudburst, deluge,
waterspout, torrent, rain cats and dogs, rain
buckets, showery, rainy, etc.), which were
November/December 2021
carefully selected to avoid false positives, resulting
from tweets that include apparently significant
words but with senses that are not related
to raining events. The search for the information
was restricted to Spain and Spanish, so language
identification was used to discard messages written
in regional languages (e.g., Catalan, Basque,
or Galician). The crawling was activated on January
19, 2020, when the deadly storm Gloria battered
Spain. The consequences of this storm
have been dramatic in eastern Spain. For
instance, the Balearic Islands and the region
around Valencia were the hardest hit on January
20, with heavy flooding and strong winds. More
than 30 provinces in Spain were put on high alert.
Figure 2 shows the influx of messages from
Twitter, RSS, and Telegram for a period of 24 h following
the activation. The blue color shows the
tweets that include some of the aforementioned
keywords. It shows a sudden increase at the date
when the event was consolidated and had greatest
impact. Indeed, events like Storm Gloria usually
have two critical points. First, authorities
warn the population about the potential risks. At
this point, if the event is perceived as dangerous
by the population, it usually has an impact on
social networks and media. This can be seen on
the left-hand side of Figure 2. Second, activity in
social networks and newspapers is usually even
greater when the natural disaster actually occurs
and causes damage to the population. This effect
is displayed on the right-hand side of Figure 2.
Therefore, the course of the activity in social networks
and media generated by this event clearly
correlates with the usual evolution of natural disasters.
This means that people's opinions in
social networks are storytelling and the analysis
93

IEEE Consumer Electronics Magazine - November/December 2021

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