164
X. Jin
Although Twitter provides real-time data for researchers to study disaster risk
communication, its word limits prevent researchers from accurately interpreting short
posts (Tang et al. 2014). Previous research has applied and extended the LDA model in
social media contexts. For instance, Hong and Davison (2010) trained standard topic
models on micro-blog messages and compared the effectiveness of these models.
They concluded that training the topic model on aggregated social media messages
can lead to better performances of modeling social media text. With these implementation efforts, topic modeling has been seen as an appropriate tool for researchers to
detect social media patterns.
With topic model analysis, researchers can identify major discussions associated with certain events. The next section reviews how topic model analysis has
been used by researchers to unpack the patterns of social-mediated disaster and risk
communication.
9.3.2 Applying Topic Model Analysis in Disaster and Risk
Communication
Topic modeling has been used to identify health trends on social media (Asghari
et al. 2018; Paul and Dredze 2014), explore disaster and risk communication patterns
(Sadri et al. 2018), and assess disaster damage (Resch et al. 2018). For instance, to
identify popular topics about Hurricane Sandy on social media, Sadri et al. (2018)
conducted a topic model analysis with 763,000 English tweets posted by the top
4,029 users from October 14 to November 12, 2012. According to Sadri et al., storm
prediction, storm watch, and preparedness were the major topics in the warning
phase. In the response phase, the major concerns were about gas/fuel, food/water,
and significant power outage. Nevertheless, Sadri et al. did not include tweets posted
by users who did not tweet about Hurricane Sandy over 100 times. Such an approach
may omit the opinions of non-frequent Twitter users including accounts of emergency
management agencies, and these results may be exaggerated.
Another example of topic modeling’s application in disaster and risk research is
Resch et al.’s (2018) study that utilized LDA to analyze the tweets related to the
Napa earthquake in August 2014. With this approach, Resch et al. detected earthquake footprints and generated a damage assessment mapping significant loss. Resch
et al.’s findings largely focused on the general topics related to disaster losses, while
open discussions about emergency management effectiveness, responsibility, and
responses to disasters have not articulated. To aid disaster and emergency management, this chapter concentrates on disaster and risk communication that emerged
from the resolution stage.
This chapter aims to bring topic modeling and latent semantic analysis to understand the complex disaster and risk communication patterns; as such, the chapter
includes a case study of disaster and risk communication related to Hurricane Maria.
In this case study, the author collected real-time data from Twitter with NodeXL
X. Jin
Although Twitter provides real-time data for researchers to study disaster risk
communication, its word limits prevent researchers from accurately interpreting short
posts (Tang et al. 2014). Previous research has applied and extended the LDA model in
social media contexts. For instance, Hong and Davison (2010) trained standard topic
models on micro-blog messages and compared the effectiveness of these models.
They concluded that training the topic model on aggregated social media messages
can lead to better performances of modeling social media text. With these implementation efforts, topic modeling has been seen as an appropriate tool for researchers to
detect social media patterns.
With topic model analysis, researchers can identify major discussions associated with certain events. The next section reviews how topic model analysis has
been used by researchers to unpack the patterns of social-mediated disaster and risk
communication.
9.3.2 Applying Topic Model Analysis in Disaster and Risk
Communication
Topic modeling has been used to identify health trends on social media (Asghari
et al. 2018; Paul and Dredze 2014), explore disaster and risk communication patterns
(Sadri et al. 2018), and assess disaster damage (Resch et al. 2018). For instance, to
identify popular topics about Hurricane Sandy on social media, Sadri et al. (2018)
conducted a topic model analysis with 763,000 English tweets posted by the top
4,029 users from October 14 to November 12, 2012. According to Sadri et al., storm
prediction, storm watch, and preparedness were the major topics in the warning
phase. In the response phase, the major concerns were about gas/fuel, food/water,
and significant power outage. Nevertheless, Sadri et al. did not include tweets posted
by users who did not tweet about Hurricane Sandy over 100 times. Such an approach
may omit the opinions of non-frequent Twitter users including accounts of emergency
management agencies, and these results may be exaggerated.
Another example of topic modeling’s application in disaster and risk research is
Resch et al.’s (2018) study that utilized LDA to analyze the tweets related to the
Napa earthquake in August 2014. With this approach, Resch et al. detected earthquake footprints and generated a damage assessment mapping significant loss. Resch
et al.’s findings largely focused on the general topics related to disaster losses, while
open discussions about emergency management effectiveness, responsibility, and
responses to disasters have not articulated. To aid disaster and emergency management, this chapter concentrates on disaster and risk communication that emerged
from the resolution stage.
This chapter aims to bring topic modeling and latent semantic analysis to understand the complex disaster and risk communication patterns; as such, the chapter
includes a case study of disaster and risk communication related to Hurricane Maria.
In this case study, the author collected real-time data from Twitter with NodeXL