1 Introduction

Disaster emergency management is a crucial component of achieving sustainable socioeconomic development because it helps to reduce the impact of disasters on the social-ecological system and is a crucial part of constructing resilient cities and society (Shi et al. 2006; Janak and Suman 2018). Social response to emergency is an important mechanism for implementing and guaranteeing emergency management of major natural hazard-related disasters (Ma 2012). Disaster information dissemination is important in regulating the process of disaster emergency management and can have an impact on how a disaster event evolves within the social-ecological system (He 2021). As global warming continues, the adverse effects of extreme weather events will become more widespread (IPCC 2021). China is a sensitive and significant area of global climate change, and as global warming intensifies, extreme climate events in the region will tend to intensify, as the level of climate risk is on the rise (Climate Center of CMA 2022). Human societies are vulnerable when extreme low-temperature events occur as a result of the global warming trend (Stanišić et al. 2016). Through cascade coupling, an extreme low-temperature event can evolve into a compound event of a sudden disaster (Ebi et al. 2021). Therefore, managing emergency responses to extreme climate events under the impact of climate change shocks to strengthen the resilience of social-ecological systems is an urgent scientific problem (Ouyang et al. 2022; UNDRR 2022).

Previous research on disaster information dissemination has focused mostly on crisis communication and reporting of emergencies, primarily in relation to social security emergencies and public health emergencies and less on major natural hazards and disasters. Studies on crisis and disaster information dissemination applications have mainly analyzed crisis and disaster information dissemination patterns, dissemination paths and information sources, hotspots, and expression methods (Havas et al. 2017; Chen et al. 2020; Babvey et al. 2021); the relationship between the government, the media, and the public in crisis and disaster information dissemination and the information-seeking behavior of the public (Leykin et al. 2016; Magee et al. 2020); social network crisis and disaster event information dissemination models (Wan et al. 2022); and disaster information distortion, such as rumors (Xie et al. 2020; Pröllochs et al. 2021).

Although there is still a lack of extraction and deep mining of crisis or disaster information content, and the processes and mechanisms of how information dissemination influences or regulates the evolutionary process of disaster events have not been clarified, studies have revealed that elements of disaster information dissemination can have an impact on the disaster or crisis behavior of the public. This influence exists at not only the community level but also the individual level (Austin et al. 2012; Freberg 2012; Liu et al. 2016). Specifically, the interaction between the credibility of disaster information sources and information quality has a significant positive effect on people’s willingness to engage in online altruistic behavior (Arazy and Kopak 2011; Lemieux 2014; Li et al. 2015), and information disseminated through social media affects the response and reaction of volunteer groups and relief organizations (Cho and Park 2013; Bruns and Burgess 2014; Oliver and Anna 2016). Additionally, information behavior that carries various sentiments accelerates the dissemination of crisis or disaster information, and through the feedback function of social media, ideas and opinions are collected from community groups and the public who may be affected by the disaster or crisis events, which affects the subsequent information behavior (Stieglitz and Linh 2013; Oliver and Anna 2016; Qiu and Ge 2020; Ran and Chen 2021). The form, source, and delivery of crisis or disaster information may potentially play a role in the public’s information seeking and sharing during a crisis or disaster, which may in turn influence people’s personal thoughts, sentiments, and subsequent disaster emergency behaviors (Liu et al. 2015, 2016).

These studies have further shown that information dissemination may be a catalyst that influences the evolutionary process of disaster events in social-ecological systems. China experienced severe freezing-rain and snowstorm disasters in 2008 that affected 21 provinces (including autonomous regions and municipalities), resulting in direct economic losses of CNY 151.65 billion (USD 21.11 billion) (Expert Committee of China National Disaster Reduction Committee 2008). Since China entered the era of network communication, this occurrence was the first catastrophic case of an extreme low-temperature incident, making it very representative. Therefore, based on the case of the 2008 southern China freezing-rain and snowstorm disasters, this study attempted to answer two research questions after processing and analyzing the components of media information during the period of the case study as it occurred and evolved:

  1. (1)

    What was covered by the dissemination of disaster information among the media during the freezing-rain and snowstorm disaster outbreak?

  2. (2)

    How did the content of disaster media information influence social response to the emergency?

2 Methodology

This section describes the conceptual design, case study details, statistical analysis methods, and data acquisition and filtering, as well as algorithms and processes for extracting hidden information from media reports.

2.1 Conceptual Design

Human decision-making behavior in human-environment interactions is determined not by the real world but rather by the perceived world that is based on the real world and viewed through a combination of various social filters, such as social media, the government, and scientists, as well as individual structural-functional filters, such as gender, age, and place of residence (Spencer and Blades 1986). The two most significant behavioral processes in geography are decision making and choice making, and information is critical in promoting the emergence of behaviors (Louviere and Henley 1977). The information that exists in society at the time of a disaster event may include information that people recall from their personal memories, information from external sources (for example, mass media), and information that people communicate and exchange (for example, by telephone, e-mail, or social media).

Information about disasters is first prefiltered by a person’s senses, conductive experience, social media, the government, scientists, and so on before entering the human brain through instinctive and mental modes. The human brain then processes the information to produce spatial cognition, after which it develops decision-making and choice behavior. The content of all types of information may significantly influence people’s emotions, attitudes, preferences, perceptions, and so on and then serve as the basis for decision making, which is ultimately manifested in disaster response behaviors.

2.2 Case Study Selection

Severe freezing-rain and snowstorm disasters swept across most areas of southern China between 10 January and 2 February 2008 (Fig. 1). The Yangtze River Basin experienced its worst extreme freezing weather in the last hundred years, while Guizhou and Hunan Provinces set records for the highest number of freezing days. Several areas had temperatures more than 4 ℃ lower than in previous years, with Guizhou experiencing the thickest ice on the ground recorded (Zhou 2008; Shao et al. 2011).

Fig. 1
figure 1

The 11 provinces that were seriously affected by the 2008 freezing-rain and snowstorm disasters in southern China (case study example locations in red)

Full size image

This event caused 129 deaths, emergency relocation of 1.66 million people, and direct economic losses of CNY 151.65 billion (USD 21.11 billion). These losses were the most severe in the history of similar disasters, affecting a large number of people. The freezing-rain and snowstorm disasters in 2008 had a profound impact on the regional and national economy and society. Typical damage examples include electricity and water outages in southeastern Guizhou, electricity and communication outages in Chenzhou City of Hunan Province, and stranded passengers at Guangzhou Railway Station in Guangdong Province.

2.3 Research Data

Before the disaster began on 10 January 2008, meteorological agencies issued early warnings, and after the freezing-rain and snowstorm disasters had started, emergency management agencies communicated disaster relief and reconstruction information to the public. Therefore, the timespan of the required data for our research was set from 1 January to 1 March 2008.

Media information on the disasters was obtained from the People’s Daily, an authoritative and influential newspaper in China, and Baidu, a Chinese full-text big data search engine.Footnote 1

Social responses to emergency data were obtained from the China Meteorological Disaster Yearbook (2009)Footnote 2 and the Emergency Plan, Emergency Resource Database of the Ministry of Emergency Management of the People’s Republic of ChinaFootnote 3 as well as case-related media reports from the Internet (Baidu) and newspapers (People’s Daily).

2.4 Data Acquisition and Filtering

We collected 7,857 media reports from paper and Internet media (Fig. 2):

  1. (1)

    We manually extracted 768 valid case-related media reports from various pages of the People’s Daily.

  2. (2)

    We used the Baidu API to extract 7,089 valid case-related media reports. The specific steps were as follows.

Fig. 2
figure 2

Flowchart of the steps and procedures for media report mining. LDA = Latent Dirichlet allocation.

Full size image

Step 1: We set the following search criteria through Google TrendsFootnote 4 to identify the case search hot-topic words with the search hot value, specifically: “1 January–1 March 2008” was the search period; “China” was the search area; and “snowstorm” was the trigger word. The final hot-topic words obtained were “snowstorm,” “freezing rain,” “Spring Festival,” “electric power system,” “highway,” “donation,” and “de-icing,” among others.

Step 2: We used the case hot-topic words as crawling keywords, set the crawl conditions, and used a web crawler to extract case-related media reports from Baidu. For full access to data, we used the hot-topic words identified in step 1 as supplementary crawling data and superimposed them on the crawling keywords one by one (example: “2008 southern snowstorm” and “hot-topic words”). Then, we again followed the same crawling procedure to roughly obtain the crawling results of the case-related data.

Step 3: The drop_duplicates () function was used to remove duplicates from the text data and obtain the final valid media report data.

2.5 Hidden Information Extraction from Media Reports

We used natural language processing (NLP) integrated with machine learning to extract hidden information from the 7857 case-related media reports (Fig. 2).

2.5.1 Attribute Word and Behavioral Word Extraction Based on the Jieba Tool and Latent Dirichlet Allocation (LDA) Algorithm

Since there are no clear word separation markers in Chinese, the language has a complicated structure, and the same word can have various meanings depending on the context (Wang and Qian 2008). Thus, tokenization must be used to divide the text into meaningful tokens. This serves as the foundation and precursor for NLP and aids in the creation of vocabularies and document vectors. Currently, the mainstream Chinese tokenization tools are Jieba, LTP, SnowNLP, and THULAC. Among them, the Jieba tool is a Python-based third-party word segmentation database. The Jieba algorithm combines the methods of string matching and statistical matching, so it is easy to use and widely applied (Wang et al. 2015).

In this study, the open-source library Jieba toolFootnote 5 in Python was used for tokenization of the 7,857 media reports before we extracted the topics and determined the attributes of the text. First, we manually selected words describing disaster impacts, disaster responses, and so on, from the original text and added them to the customized user dictionary through the Load_userdict () function in the Jieba library in order to improve the precision of tokenization. This operation avoided tokenizing the words “warning issued (发布预警)” into “warning (预警)” and “issued (发布)”, “house collapse (房屋倒塌)” into “house (房屋)” and “collapse (倒塌)”, and so on. Then, we used the Remove.stopwords () function in Python’s scikit-learn library and referred to the Baidu Stop-Word List, SCU Stop-Word List, and HIT Stop-Word ListFootnote 6 to remove prepositions, modal particles, punctuation, special symbols (that is, @, %, #), and words with no practical significance, such as “的,” “和,” and “是.” We checked all tokenization results and made timely human corrections to inaccurate software analysis results to ensure that the subsequent text classification was accurate.

We transformed the media report text after the completion of tokenization into an iterator of SENTENCE as the input document for the model and called on the word2vec built into Python’s Gensim library for training to generate the LDA (latent Dirichlet allocation) topic model for recognizing the topics of case-related media reports. The LDA has three levels—word, topic, and document (Onan et al. 2016); that is, it uses an unsupervised machine learning algorithm to map the document onto the topic and can dig deeper into the hidden information in the text (Albalawi et al. 2020).

The Dirichlet distribution parameters α and β for the document-topic and topic-word distributions, as well as the number of topics (K), must be determined before LDA. The LDA operating concept states that for the nth word in a document, the subject must first be extracted from the topic distribution of the document, and the word must then be extracted from the word distribution corresponding to this topic. This random process is continued until all of the words in the document have been extracted. The joint probability of each word appearing in a document can be represented as (Blei et al. 2003):

$$P\left( {\theta ,t,w\alpha ,\beta } \right) = P\left( {\theta \alpha } \right)\mathop \prod \limits_{n = 1}^{N} P\left( {t\theta } \right) \times P\left( {wt,\beta } \right)$$
(1)

where t represents the topic, w is the word in the document, α is the topic probability distribution, and β is the probability distribution of word items.

In this study, we used Gibbs sampling to train the model, which repeats sampling until convergence and automatically learns the ideal parameters α and β. We considered the experimental parameters mentioned by Liu and Hu (2012) and set the value interval of the number of topics to [5,20], and the maximum number of iterations to 50 by default. According to the trend of perplexity, 9 types of topics (K = 9) were finally extracted and artificially classified into the hazard-affected elements (Topic 4 to Topic 9), disaster impact (Topic 1) and disaster response (Topic 2 to Topic 3) (Fig. 3).

Fig. 3
figure 3

Attribute word and behavioral word extraction results

Full size image

2.5.2 Extension of Attribute Words and Behavioral Words Based on the Random Forest Algorithm

Machine learning algorithms for text categorization include K-nearest neighbor (KNN), logistic regression, decision tree, support vector machines (SVMs) (Breiman 2001), and others. However, ensemble learning with multiple learners has been shown to yield better results than a single learner (Shen et al. 2022).

The random forest algorithm is a combinatorial classification prediction algorithm proposed on the basis of the bagging algorithm and ensemble learning (Gieseke and Ige 2018). It uses multiple decision trees for supervised regression classification (Breiman 2001). For each node of the base decision tree, if the total number of attributes of the node is d, a subset of k (usually log2 d) is randomly selected from all the d attribute features, and the advantages and disadvantages are then compared in this subset to obtain the optimal partition attribute. After training classification model m {h1 (x),h2 (x),...,hm (x )}, the output of the final random forest classifier H (x) is the vote in the output of this m classification mode (Breiman 2001):

$$H\left( x \right) = \arg max\mathop \sum \limits_{i = 1}^{m} I\left( {h_{i} \left( x \right) = Y} \right)$$
(2)

where H (x) is the last classifier generated for the random forest and Y is the output variable.

We regressed the media reports with the random forest algorithm as the training classifier for the extension of attribute words and behavioral words and obtained the classification judgment words tree of the hazard-affected elements (Fig. 4) and the classification judgment words list of the impact and response (Fig. 5). Topic 1 (disaster impact), Topic 2–Topic 3 (disaster response), and Topic 4–Topic 9 (hazard-affected elements) were selected as the training set.

Fig. 4
figure 4

Judgment words tree of the hazard-affected elements

Full size image
Fig. 5
figure 5

Judgment words tree of the impact and response

Full size image

2.5.3 Judgment of the Attributes of Media Reports According to the Words Tree

Based on Figs. 4 and 5, the 7,857 media reports were artificially discriminated on an attribute-by-attribute basis. When discriminating, three situations were handled as follows:

(1) where the same report had more than one hazard-affected element, each hazard-affected element was counted separately; (2) where the same report had both the disaster impact and the disaster response information, they were counted separately; and (3) where multiple words in the same report described a disaster impact or disaster response or hazard-affected elements, they were counted only once.

For example, one report stated that “Affected by the ice-snow disaster once in half a century, the high-voltage line of the railroad in the Hunan section of the Beijing-Guangzhou line was covered by ice, causing train delays. The Guangzhou Railway Group quickly organized more than 1000 repair team members together with the local power supply departments to carry out electric power and railway repair, with tens of thousands of employees in more than 100 stations along the line to strengthen the safety inspection of the obstructed trains.” When making judgments, “line was covered by ice” and “train delays” as impacts of the disaster on the hazard-affected elements electrical facilities and transportation, respectively, were handled as described in (1) above; “train delays” and “strengthen the safety inspection of the obstructed trains” as descriptions of impacts and responses to the same hazard-affected element, transportation, were handled as described in (2) above; “strengthen the safety inspection of the obstructed trains” and “railway repair,” both as part of the response to the transportation problem, were handled as described in (3) above.

2.6 Statistical Analysis Methods

Grounded theory (GT) is a qualitative research method that builds theory from observations. It starts with no assumptions and collects data to reflect the essence of the phenomenon, forming a systematic theory (Strauss 1987). In this study, GT was used to summarize the media information content, and analyze media reports and their role in social responses to emergencies.

Granger causality analysis (GCA) is a method for determining cause-and-effect relationships. It quantitatively validates relationships between variables (Russo 2009). Granger causality analysis does not establish causality, but it supports causal mechanisms based on statistical findings (Zhang et al. 2011). This study examined the relationship between media reports and social response to emergencies. Before GCA, each variable underwent an augmented Dickey-Fuller (ADF) test to assess the smoothness of the time series data. The widely accepted Schwarz information criterion (SIC) was used as the statistical criteria for time delay indices.

3 Results

This section presents the findings based on hidden information extraction from media reports, including the main content of media reports during the disasters and the relationship between social response and media reports.

3.1 Identification of the Main Content of Media Reports during the Freezing-Rain and Snowstorm Disasters

As shown in Table 1, the content of media reports was analyzed text by text. After identification and summarization, the disaster reports were subdivided into nine categories, with those containing the disaster impact information categorized as 1, disaster real-time status and the disaster response information was subdivided into eight additional categories (2–9) (Table 1).

Table 1 Classification and description of the case media report content
Full size table

3.2 Analysis of the Relationship between Media Reports and Social Response to Emergencies

Three typical case study examples were extracted after discriminative media report content analysis to identify the relationship between social response and media reports.

3.2.1 Case Example I of Electricity and Water Outages in Southeast Guizhou

Figure 6 displays a timeline of actions of various actors in this specific case example, extracted from the 462 case-related media reports.

Fig. 6
figure 6

Response process in case example I: electricity and water outages in Southeast Guizhou. Solid orange boxes are for disaster real-time status (category 1) and social mobilization (category 4); solid blue boxes are for response real-time status—donation behavior (category 5)

Full size image

As shown in Fig. 6 reports in the disaster real-time status and social mobilization categories appeared before reports in response real-time status—donation behavior. The correlation analysis results (Table 2) show that media reports of response real-time status—donation behavior were significantly positively correlated with disaster real-time status and social mobilization (p < 0.01), and social mobilization was also significantly positively correlated with disaster real-time status. On the basis of the correlation analysis, the cause‐and‐effect relationship between media reports and social response was further validated. The smoothness of the time series for disaster real-time status, social mobilization, and response real-time status—donation behavior passed the ADF test. The GCA results (Table 3) show that the relationships between response real-time status—donation behavior, disaster real-time status, and social mobilization all passed the Granger causality test at the 0.01 significance level under the SIC. This demonstrates statistical cause‐and‐effect relationships between the variables.

Table 2 Correlation coefficient of media reports content
Full size table
Table 3 Granger causality analysis (GCA) of media reports content
Full size table

Moderate media reports of disaster situations can induce a strong psychological and cognitive sense of crisis and urgency in the public, thus creating an incentive effect. After the government and other social organizations appealed to all parties to fight the disasters, other actors reacted accordingly to participate in disaster relief. Consequently, the dissemination of media reports on disaster real-time status and social mobilization stimulated social responses.

3.2.2 Case Example II of Electricity and Communication Outages in Chenzhou City, Hunan Province

Figure 7 displays a timeline of the evolution of this case example in the social system, extracted from the 439 case-related media reports. The first report appeared on 22 January when the Central Meteorological Station issued a yellow warning for freezing (category 2, prewarning and forecasting) before each weather report to remind people to avoid risks. However, there were electricity and communication outages in the disaster area as the disaster situation worsened. Social responses—such as donating radios and batteries, increasing the frequency and spread of newspaper delivery, and starting a disaster relief radio channel—all occurred in order to maintain communication with people in the disaster area in addition to accelerating the repair of electrical towers and communication base stations. Additionally, the provincial broadcasting station steadily provided positive reports of anti-ice relief efforts in Chenzhou City to meet the information needs of individuals in the disaster area and to highlight those who made significant contributions (category 7, empathetic care and category 8, relief achievements recorded) to internalize people’s emotions, attitudes, and values. The media also provided timely information about the disaster and urgent relief needs in the disaster area (category 1, disaster real-time status and category 4, social mobilization).

Fig. 7
figure 7

Timeline of the case example II events: electricity and communication outages in Chenzhou City, Hunan Province. Solid orange boxes are for disaster real-time status (category 1), prewarning and forecasting (category 2), announcement by the authority (category 3), social mobilization (category 4), dissemination of scientific knowledge (category 6), empathetic care (category 7), and relief achievements recorded (category 8); solid blue boxes are for response real-time status—donation behavior (category 5)

Full size image

As a result, the media’s involvement in disseminating disaster information alerted the public, minimizing risks and losses. It eliminated “information islands” and maintained social stability by ensuring information exchange between disaster and nondisaster areas, the government, and residents. This multisectoral cooperation lessened the harm caused by the disaster events through the coordination and interaction of all forces.

3.2.3 Case Example III of Passengers Stranded at Guangzhou Railway Station, Guangdong Province

Figure 8 displays a timeline of the daily variations in the number of stranded passengers in Guangzhou Railway Station in this specific case example, extracted from the 286 case-related media reports.

Fig. 8
figure 8

Timeline of case example III: passengers stranded at Guangzhou Railway Station, Guangdong Province. The red dashed box shows examples of conflicting information; orange solid boxes are reasons for the surge in stranded passengers; blue solid boxes are reasons for the decrease in the number of stranded passengers

Full size image

Figure 8 shows two spikes in the number of stranded passengers, 29 January and 1 February. Beyond the Beijing-Guangzhou line closure due to snowfall, decisions made by passengers were influenced by news of ticket sale suspension, airport opening, line restoration, and the government’s New Year call to stay local. This shows that media reports can influence individual decision making. Combined with behavioral theory, these results reveal that risk information affects behavior and decisions through systematic and heuristic information processing (Trisolini et al. 2004). The public typically follows a heuristic information processing method that involves analyzing incoming information using straightforward decision criteria, overlooking some perceptions, and making decisions or acting in certain ways immediately (Paul et al. 2020). Individuals choose information relevant to their interests frequently, particularly during unpredictable disasters (Hall et al. 2021). In addition, because the media dispelled rumors about the Guangzhou Railway Station shutdown, the number of people who flocked to the railway station as a result of rumors was reduced to some extent. This illustrates how the media can direct public opinion and eliminate negative public opinion and how the dissemination of accurate disaster information plays a role in maintaining social order.

4 Discussion

This section discusses the driving mechanism and information requirements of social response, as well as the differences in the impact of disaster information sentiment on public sentiment about disaster emergency response, and gives future research directions based on the limitations of this work.

4.1 Empathy as One of the Drivers that Stimulates Social Response to Emergencies

We analyzed the relationship between social response and media reports through three typical cases. It is obvious that media reports can influence how various actors react to disaster events. Most actors work in nondisaster areas, but they changed from bystanders to “disaster responders” who actively took part in relief efforts. The formation of empathy motivated their psychological and behavioral identity transition in reaction to the disasters.

According to psychological theory, donation is a form of autonomous, spontaneous, and voluntary prosocial activity, and empathy, moral judgment, and perspective discrimination are the driving forces behind it (Christner et al. 2022). Empathy is an alternative emotional response in which the observer experiences the same emotions as the observed person when perceiving the environment or emotional state of the observed person, and emotional empathy is a necessary precondition for prosocial activity (Rieffe et al. 2021).

In disaster information dissemination, media reports present the disaster situation and relief progress in multiple authentic and timely forms. The objective description of the disaster case facts prompts various actors to develop empathy, which in turn causes them to react to the disaster event and generate a series of autonomous, spontaneous, and voluntary donation behaviors. The widespread dissemination of media information on the disasters also brings together all segments of society to establish a social synergy to jointly respond to the crisis, which in turn brings together the response behaviors of these actors. Additionally, media reports on donation behavior (category 5 response real-time status and category 8 relief achievements recorded) serve as positive demonstrations and examples, uniting people’s emotions and inspiring the public. This stimulates inherent empathy, leading to “second-order empathy” where people in nondisaster areas share the loss of those affected (Dillard 1986). However, the quantity and content of disaster real-time status should be carefully controlled. Excessive exposure to this information can evoke negative sentiments like panic, anxiety, and anger (Li et al. 2022).

4.2 Differences and Stages in the Information Requirements of Social Response to Emergencies

The analysis of three typical case examples shows that information requirements differ at different stages of disaster evolution. Information requirements arise when an unforeseen disaster occurs, and the public realizes that their knowledge is insufficient to deal with the abnormal occurrence. At this point, the public’s information requirements diverge from the routine, and they will more frequently use social media to repost disaster information (Thelwall and Stuart 2007).

The process of disseminating information about disasters follows the same life cycle as the development of disaster situations. From before the occurrence of meteorological disaster events to the latent phase, the intensity of hazard factors and the disaster-causing ability and influence range are limited. At this time, the information requirements of the society cover categories 1, 2, 3, 5, and the core of category 2 prewarning and forecasting to maximize disaster risk avoidance and prevent potential risks or possibilities of disaster events. During the disaster outbreak phase, numerous negative effects, including traffic disruption, power and communication outages, and energy supply shortages, appear in a chain, simultaneously creating a vicious cycle through the cascade coupling of each affected element. At this stage, society’s information needs include categories 1, 2, 3, 5, 6, 7, and 8 as well as the core of categories 4 social mobilization and 5 response real-time status. While actively influencing public opinion and reducing the spread of rumors, the media must, on the one hand, provide accurate, timely, and continuously updated reports on the facts of the disasters. On the other hand, they should exploit multiple media to guide the public to consume information to form a social synergy in response to the disasters and recover as quickly as possible. Information requirements of the public during a disaster’s recovery phase in categories 3, 4, 5, 6, 7, 8, and 9 serve as the overarching theme, as does the core of category 9 summarization and reflection. To achieve the goals of reconstruction and the restoration of social order in disaster areas, the media must, on the one hand, continue to track the development of post-disaster reconstruction work, supervise public opinion, and ensure the scientific and orderly development of restoration and reconstruction work in disaster areas. On the other hand, the national emergency management department should review and consider common issues that emerged during the occurrence and emergency response to a disaster event as well as the lessons learned and the anticipated future response to such events in the hope that the anticipated response will be proactive and increase the resilience of the society.

4.3 Differences in the Impact of Disaster Information Sentiment on Public Sentiment about Disaster Emergency Response

In terms of sentiment attributes, disaster information content can be divided into two categories—positive and negative. It has been proven that sentimental disaster information in the media inspires actions (Huang et al. 2021). According to the emotional infection theory, images, videos, texts, and other information relating to events trigger and affect the audience’s sentiment, which has a mediating influence on the transmission of other sorts of information (Choi and Toma 2014; Brierley et al. 2022). When severe sudden natural hazards and disasters happen, in particular, their distinct and unique emergent characteristics stimulate the dissemination of a large amount of sentimental information. This sentimental information awakens people’s intrinsic physiological mechanisms and unconsciously produces the corresponding sentiment, which in turn stimulates them to exhibit behaviors that are in line with their emotions (Anna 2009; Stieglitz and Linh 2013).

It has been demonstrated that the dissemination of positive information fosters the development of positive attitudes among people and affects groups through emotional infection mechanisms, which in turn encourage cooperative group behavior (Liu and Liu 2013). Categories 2, 3, 4, 5, 6, 7, 8, and 9 are positive information about a disaster (see Table 1). Specifically, the government introduces the relief measures that are being taken for people in the disaster area through newspapers and broadcasts and provides timely information about the progress of rescue. This not only propagates positive emotion and can dissolve dissatisfaction with the government fostered by the environment of the disaster area where people have been for a long time but also relieves the tension of people in the disaster area. Various government departments, such as meteorology and transportation, provide timely road condition information and meteorological forecasts through the media. This information can be of assistance to local businesses and individuals in reducing production losses and loss of life. In contrast, the excessive negative information provided by disaster real-time status causes unfavorable strong public sentiment reactions, such as panic, anxiety, and anger (Li and Wang 2022), thereby amplifying the event’s adverse effects through intense informational behavior, creating a gathering effect, and promoting group conflict behavior. This gradually breeds substantial dissatisfaction with the government and may lead to social instability, and causes natural hazards and disasters to become social crises.

Information inconsistencies can lead to rumors through emotional infections to influence individual decisions and negatively affect disaster preparedness and response. In the case of passengers stranded at Guangzhou Railway Station (see Fig. 8), the central government and the Ministry of Railways were more optimistic about the railroad’s recovery and promised to send the passengers on their way before the Spring Festival, but the Guangdong provincial government had been pleading, encouraging, and advocating for migrant workers and passengers to stay in Guangzhou for the holiday since 28 January. The Guangdong government released its information earlier than the national government. Disaster emergency response involves subjective elements and various interest groups. The government, as the primary information source, may ignore certain details to maintain a public image (Mon 2009). This can create gaps between the government and functional organizations, as well as the public, causing information roadblocks, delays, and inconsistencies.

The timely, accurate, and effective dissemination of authoritative prewarning, disaster status, and rescue progress information is directly related to the public’s sense of security as well as social stability and sustainable development in natural hazards and disasters with sudden, complex, destructive, urgent, and unpredictable characteristics.

We did not distinguish other aspects of media reports, such as speed, source, and dissemination environment, and the effects of these aspects on social response to disasters are not included in this study. Likewise, mechanisms of media reports content regulating the evolutionary processes of disaster events in social systems were also not examined. These two issues will be the focus of follow‐up studies. Information dissemination that regulates the entry-exit transition of disaster events in social systems through various social media also requires further investigation.

5 Conclusion

In this study, based on the case of the 2008 southern China freezing-rain and snowstorm disasters, the content of media reports was identified, and the relationship between media reports and social response to emergency was analyzed empirically. The study’s four main conclusions are:

  1. (1)

    The prewarning and forecasting information in the media guarantees that appropriate disaster defense deployment is made immediately, avoiding risks in advance and effectively contributing to the prevention and reduction of disaster-related losses.

  2. (2)

    Individual decision making is influenced by media information on disaster real-time status, social mobilization, and response real-time status, which causes bystanders to become disaster responders through the empathy mechanism. These categories of media information guide various actors to actively take part in relief and rescue efforts, forming social synergy among all segments of society in responding to disaster events and encouraging more actors to participate in disaster relief through positive demonstration and examples. It guarantees the efficient execution of emergency response to disasters and contributes to the well-organized execution of disaster relief efforts.

  3. (3)

    Media information on disaster real-time status (appropriate amounts), announcements by the authorities, response real-time status, dissemination of scientific knowledge, empathetic care, and relief achievements recorded objectively describe the facts of the disaster event. If this information is made widely available, “information islands” can be eliminated, and the government, residents in the disaster area, and the general public who are concerned about the crisis can all receive pertinent information. It also facilitates information exchange between disaster-affected and unaffected areas, effectively addresses social crises brought on by natural hazard-related disasters, ensures effective post-disaster rehabilitation and reconstruction, and promotes mental stability and social order.

  4. (4)

    Disaster real-time status information (excessive amounts) and “noise flow” caused by distorted and unequal media information may mislead public opinion and response, affect individual decision making, and negatively affect the development of disaster events, increasing the unpredictability of their development and the risk of social crisis.