Analyzing Social Media Users’ Information Reposting Behavior after Disasters

社交網絡用戶的災情信息轉發行為研究

Student thesis: Doctoral Thesis

View graph of relations

Author(s)

Related Research Unit(s)

Detail(s)

Awarding Institution
Supervisors/Advisors
Award date23 Jan 2019

Abstract

After disasters, many types of disaster-related information are spreading and being exchanged on social media platforms. Some information is useful for the timely understanding of the disaster and post-disaster recovery, thus enhances facilitates the emergency and post-disaster management. Some information, on the other hand, is not useful (or even harmful) for emergency and post-disaster management. From the perspective of the government, understanding the dissemination process of various types of disaster-related information in social networks, and identifying the factors that affect the various types of information dissemination process, can provide a basis for the government to monitor such information propagation patterns, and to publish disaster-related information accordingly. Thus, this research combined text mining and social network analysis techniques to characterize, model, and predict the dynamic propagation process of various types of information after natural disasters by harnessing a unique Sina Weibo data for an earthquake occurred in Yiliang, China, in 2012.

First, we research on identifying types from social media discussions and characterize the propagation patterns of different types of information. In the past, only the classification of disaster information was realized, or only the propagation network of disaster information was considered (unclassified). However, research on the dynamic propagation patterns of different types of information after disasters is rare. To address this challenge, the information reposting behavior of users was analyzed by combining classification and social network analysis methods. And the frequency of the two interaction modes (cascade mode and reciprocal mode) appearing in the information forwarding network is calculated. Combined with the social exchange theory, the user's information demand and engine are explained. The analysis of network attributes and interaction patterns shows that: in the early stage of the earthquake, social network users mainly reposted the Casualties and damages of the earthquake directly, so such information is easily overloaded; in the late stages of the earthquake, social network users will mainly discuss the Donation of money, goods or services information with each other; Seeking for Help, Caution and advice types of information have not widely disseminated, thus, it is necessary to adjust its information release strategy, increase its dissemination scale, and thus improve the efficiency of emergency response. In addition, the method innovation of this thesis is that the comprehensive text classification and social network analysis methods, which is worthy of reference by researchers in other fields.

Second, we analyzed the moderating effect of the social capital of users on the information propagation scale and depth. Previous studies have only considered the direct impact of users' social capital on the information propagation process. Based on the social capital theory, this thesis analyzes the effect of the content type, the sentiment of the content, the geographic location of the user, and the direct and moderating effect of the social capital of users on the propagation scale and depth of the disaster-related information. The negative binomial regression results show that: a) the information published by high social capital users near the disaster area is larger and deeper; b) the information sent by high social capital users expressing positive emotions are more likely to cause large-scale reposting; c) The disaster reporting information, donation materials, and questioning information are more deeply spread; the disaster reporting information released by high-activity users can cause large Casualties and Damages, Donation of Money, Goods or Services and Doubts from ordinary Weibo users can spread deeper. The emergency management department can apply the network modeling, text classification, and regression analysis methods of this thesis to formulate specific information release strategies for different types of disaster-related information. The theoretical innovation of this study is to expand the application scope of social capital theory and examine the moderating effect of users' social capital on other factors.

Third, the role of opinion leaders in the dissemination of different types of disaster-related information was studied. Although previous studies have emphasized the importance of opinion leaders in the process of information dissemination, they have not considered the influence of opinion leaders on different types of information dissemination processes. This study first identifies the opinion leaders in the process of disseminating different types of disaster information and finds that not only high social capital users can become opinion leaders, but whether users can become opinion leaders is highly correlated with the category of information they publish. Secondly, combining the two-step flow of communication theory with the social capital theory, it analyzes the influence of opinion leaders on the propagation scale, the speed of disaster-related information, and the influence of opinion leaders on public opinion when sharing different types of disaster-related information. The results of linear regression analysis and network analysis show that the participation of opinion leaders can expand the propagation scale and speed of disaster-related information such as Donation of Money, Goods or Services, Doubts information. But it does not significantly affect the public's opinion when the public share Personal Related and Others information. Therefore, opinion leaders are less important in appease public emotions. The emergency management department can apply the research methods and conclusions of this study to set agenda for guiding the public opinion; the researchers who are committed to discovering the key nodes in the network can learn from this research to identify the key nodes based on the content to be promoted.

Fourth, this thesis proposed the method and features for predicting the propagation scale of disaster-related information and identifying the information that can lead to hot discussions. In the past, the model for predicting information dissemination only focused on the scale and speed of information dissemination and did not consider whether information could trigger hot discussions. Combining the research results of the first three studies and the homogenous theory, this study uses the supervised machine learning methods to find out the features that can accurately predict the propagation scale (represent by the occurrence of the cascading patterns) and the features that can predict the likelihood of one piece of information being hot discussed (represent by the occurrence of the reciprocal patterns). The results show that the information release time, text length, information publisher and information sharers’ features can accurately predict the extent to which information is hotly discussed; while information release time, text length, text type, and information publisher features can accurately predict the propagation of disaster-related information. The emergency management department can apply the analysis method of this study to accurately predict the scale of information dissemination or the degree of hot discussion in combination with the key features mentioned above, so as to formulate information publishing strategy in advance. In addition, the network analysis method and the prediction model establishment method of this study can also be used for new product promotion, which is convenient for product publishers to understand the attitude of social network users to products.

    Research areas

  • Disaster information propagation, Social media users, Information reposting patterns, Influential factors, Social network analysis