9 Understanding Social-Mediated Disaster …
163
examined partial communication patterns associated with disasters. The complete
picture of crisis information emerging from social media remains unclear.
To disentangle the conundrum, data mining tools can be utilized to analyze realtime social media messages and allow researchers to study open discussions of disasters (Kavanaugh et al. 2012). In particular, this chapter proposes to utilize topic
modeling to identify communication patterns surrounding disasters.
9.3 Topic Modeling and Social-Mediated Disaster and Risk
Research
Social media provides a large volume of real-time data, which creates both opportunities and challenges for researchers to explore how social media impacts disaster
and risk communication. On the one hand, researchers have offered some insights by
analyzing retweet frequency of various stakeholders (Lovejoy et al. 2012), exploring
the strategies to increase retweet numbers (e.g., URL and multimedia files, Lachlan
et al. 2014), or conducting a content analysis of disasters related tweets with selected
samples (Lin et al. 2016a, b; Spence et al. 2015). On the other hand, although these
studies shed light on how social media impacts communication and information
dissemination, it remains difficult to capture the chaotic disaster and risk communication patterns on social media. Because the limited sample size and traditional
approach of exploring the linear relationship and group differences (Murphy 1996)
can not fully handle the unstructured social media data. Facing this challenge, this
chapter recommends bringing big data tools in disaster and risk communication
research and particularly focuses on topic modeling.
9.3.1 Introduction of Topic Model
Topic modeling allows researchers to analyze a large volume of unstructured social
media data and interpret meaning embedded in the data. Topic modeling has been
used as a tool to detect text meaning for three decades (Rohani et al. 2016). Hofmann
(2001) introduced the Latent Semantic Analysis (LSA) model in which the order of
words is neglected while generating topics. Blei et al. (2003) developed the model of
Latent Dirichlet Allocation (LDA) to identify topics by calculating the probability
distribution over documents. A topic model in LDA specifies a simple probabilistic
procedure to generate documents on the basis of latent (random) variables (Blei et al.
2003). It is also worth noting that topic models can capture polysemy as one word
can be part of multiple topics. The goal of fitting a generative model is to find the
best set of latent (random) variables that explain the observed data (Griffiths and
Steyvers 2004; Steyvers and Griffiths 2007).
163
examined partial communication patterns associated with disasters. The complete
picture of crisis information emerging from social media remains unclear.
To disentangle the conundrum, data mining tools can be utilized to analyze realtime social media messages and allow researchers to study open discussions of disasters (Kavanaugh et al. 2012). In particular, this chapter proposes to utilize topic
modeling to identify communication patterns surrounding disasters.
9.3 Topic Modeling and Social-Mediated Disaster and Risk
Research
Social media provides a large volume of real-time data, which creates both opportunities and challenges for researchers to explore how social media impacts disaster
and risk communication. On the one hand, researchers have offered some insights by
analyzing retweet frequency of various stakeholders (Lovejoy et al. 2012), exploring
the strategies to increase retweet numbers (e.g., URL and multimedia files, Lachlan
et al. 2014), or conducting a content analysis of disasters related tweets with selected
samples (Lin et al. 2016a, b; Spence et al. 2015). On the other hand, although these
studies shed light on how social media impacts communication and information
dissemination, it remains difficult to capture the chaotic disaster and risk communication patterns on social media. Because the limited sample size and traditional
approach of exploring the linear relationship and group differences (Murphy 1996)
can not fully handle the unstructured social media data. Facing this challenge, this
chapter recommends bringing big data tools in disaster and risk communication
research and particularly focuses on topic modeling.
9.3.1 Introduction of Topic Model
Topic modeling allows researchers to analyze a large volume of unstructured social
media data and interpret meaning embedded in the data. Topic modeling has been
used as a tool to detect text meaning for three decades (Rohani et al. 2016). Hofmann
(2001) introduced the Latent Semantic Analysis (LSA) model in which the order of
words is neglected while generating topics. Blei et al. (2003) developed the model of
Latent Dirichlet Allocation (LDA) to identify topics by calculating the probability
distribution over documents. A topic model in LDA specifies a simple probabilistic
procedure to generate documents on the basis of latent (random) variables (Blei et al.
2003). It is also worth noting that topic models can capture polysemy as one word
can be part of multiple topics. The goal of fitting a generative model is to find the
best set of latent (random) variables that explain the observed data (Griffiths and
Steyvers 2004; Steyvers and Griffiths 2007).