Emotional contagion on social media and the simulation of intervention strategies after a disaster event: a modeling study

Abstract

With the advent of climate change and the 5 G era, online communities are increasingly becoming the main medium for information dissemination after emergencies such as natural disasters. The widespread dissemination of negative online information may generate cyber violence or lead to serious adverse psychological outcomes. This study considered a natural disaster event involving avoidable deaths and child casualties as an example to identify emotional contagion and conduct simulation interventions. Data about the aftermath of the 8·13 flash flood in the Longcaogou Scenic Area, Sichuan Province, China, were derived from the Chinese Sina microblog. We analyzed key parameters and modeled them in a dynamic model. We further evaluated the effects of implementing intervention measures (such as transmission path interruption and changing the number of different emotions) on emotional spread. The overall sentiment of posters after this flood was negative, with three epidemic peaks. Negative emotions were more persistent and contagious than positive emotions. Reducing the number of negative blog posts by half could have led to a 14.97% reduction in negative comments and a 7.17% reduction in positive comments. Simultaneously, reducing the number of negative blog posts and increasing the number of positive posts would have helped reduce the relative ratio of negative to positive comments. The findings have theoretical and practical implications for developing an emotional contagion model and formulating intervention strategies to guide public opinion after an emergency that involves extensive online debate.

Introduction

Climate change is producing an increasing number of extreme weather events, which have caused a surge in natural disasters over the past 50 years. Global warming may increase the frequency of flood events due to rising sea levels. There has been a recent surge of reports about natural disasters on social media since social media are popular communication tools that significantly impact people’s interactions in modern society. Social media has become an essential channel for disaster-related information and provides opportunities for the public to express their opinions and attitudes immediately through comments (Bergstrom and Wadbring, 2015; Son et al. 2019). Natural disasters have substantial mental health consequences for the public, particularly for people in disaster areas (Goldmann and Galea, 2014; North, 2016). Previous studies of social media data after natural disasters have shown that these disasters can elicit emotional reactions from the public on social media (Anderson, 2021; Garske et al. 2021). Public expressions regarding extreme weather disasters on social media were relatively emotional; they praised the good, consoled the victims, and condemned the villains (Han et al. 2022). Creating a positive online emotion-oriented environment for public opinion is a means to prevent social problems caused by the contagion of negative emotions after natural disasters and facilitate the social governance of these events.

The internet penetration rate in most countries is now extremely high. The Organization for Economic Co-operation and Development (OECD) nations average for internet use is more than 4 hours per day, and in some countries, the average is more than 6 hours per day (Max et al. 2015). According to the China Internet Network Information Center (CNNIC), by June 2022, China had 1.051 billion internet users, with an internet penetration rate of 74.4% (Central People’s Government of the People’s Republic of China, 2022). The internet penetration rate of minors in China reached 94.9%, and the age of access to the Internet is decreasing (China Internet Network Information Center, 2021). Internet use greatly impacts people’s emotions and behaviors, especially those of minors. In the absence of a friendly online environment, people may easily imitate negative behavior or generally use negative words to express emotions, which may lead to subsequent cyber violence and real-world bullying or violence (Patchin and Hinduja, 2010; Ho et al. 2020).

Sina Weibo is one of the most widely used social media platforms in China. It facilitates the dissemination of emergency information and enables netizens to access news and events and to publicly share and discuss individual attitudes, opinions, events, and activities. In addition, social media allow public health workers and emergency responders to act more quickly and efficiently after natural disasters and environmental disasters and regarding other environmental concerns. Thus, Sina Weibo data are emerging as a key online media and data source for researchers to explore social problems in a noninvasive way. This study selected an emergency involving avoidable deaths and child casualties as an example to build a model depicting emotional contagion on Sina Weibo and to simulate public opinion regulation strategies that can prevent the spread of negative emotions.

Research background and significance

Due to the development of the internet and the widespread use of social media, natural disaster events, such as floods, have sparked public discussions and exchanges of opinions. These discussions contain public emotions, which can spread through behaviors such as reposting and commenting. The process of spreading information with emotional content is referred to as emotion spread. Emotions have a substantial effect on human decision-making and behavior. In some public crises, the combination of situational triggers and cognitive biases caused by negative emotions amid the clamor of public opinion can interact and ultimately lead to a public opinion crisis, affecting social stability. Therefore, research on the emotional spread of flood events is important for social governance in the aftermath of disasters. Events related to avoidable death and children often spark more public discussion (Chu et al. 2021). This study selects a flood event that attracted public attention and generated evolving public opinions. We explore the evolution of emotional spread based on the process of information spreading.

Previous studies often focused on classifying, describing, and predicting public sentiment and analyzing the factors influencing emotions during flood events. Online emotions are infectious, fluid, and continuous (Sun. 2016). The emotions of group members can be infectious, and different types of emotions can lead public opinion in different directions. Public emotions on social media also fluctuate over time. Clarifying the transmission mechanism of emotions on social media and predicting the dynamic transmission of public emotions can provide a theoretical basis for guiding and regulating online expressions of emotion. However, existing studies have paid little attention to the patterns of emotional spread after flood events, the intensity of the spread of different emotions, and the effectiveness of different emotional management measures. Netizens’ comments and attention to event-related information promote the rapid spread of emotions in the network environment (An et al. 2021). This study aims to propose a driving mechanism for the transmission of emotions from posts to comments based on a dynamic model and to explore appropriate measures to create a better online community within the context of flood events in China. We evaluate the model-fitting effect and simulate the differences in emotional characteristics caused by different intervention measures (interrupting emotional contagion and varying the number of emotional posts). Compared to previous studies that focused mainly on emotional characteristics and complex emotional transmission relationships among users, this study can provide social media managers with more practical recommendations on public emotions. We hope to minimize the impact of floods on public mental health and help with real-time management strategies for flood mitigation.

Literature review and theoretical background

Tendency of emotional orientations on social media after a disaster event

During disasters, individuals can use social media to disseminate requests for assistance and express their sentiments. The public’s emotions fluctuate over time driven by the developments of the disaster, and these variations may be discerned through the information that individuals post online. Understanding the temporal trends in people’s emotions following sudden natural disasters, as gleaned from social media postings, can aid researchers in understanding the psychological aftermath of disasters. Numerous studies have shown that people’s emotions after natural disasters are tied to different disaster periods (Gruebner et al. 2016; Gruebner et al. 2018). For instance, some researchers analyzing the spatiotemporal distribution of negative emotions among Twitter users following flood events found that the prevalence of adverse emotions peaked during the disaster’s zenith, significantly declining in the postdisaster phase (Karmegam and Mappillairaju, 2020). Studies conducted in the Chinese context have revealed substantial public emotional responses prior to and during heavy rainfall and flooding, predominantly characterized by negative emotions. However, as relevant authorities effectively implement emergency measures, people’s emotions gradually become less intense, accompanied by frequent positive emotions on social media platforms (Ma et al. 2023). An analysis of the temporal trends of emotions among Weibo users regarding the topic “Henan rainstorm mutual aid” revealed that the largest number of Weibo messages exhibited a neutral emotional tendency, followed by messages with a negative emotional orientation. In the early stages of a disaster, the public shared and disseminated a large amount of disaster-related information, with emotions leaning toward neutrality. During the process of mutual aid and rescue, social media users displayed negative emotions. As the situation gradually improved, positive emotions such as prayers and gratitude emerged (Li et al. 2023). These studies used social media data to elucidate the relationship between disaster events and public sentiment, aiding in natural disaster management. Nevertheless, the spreading mechanism of public emotions via social media is intertwined with the propagation of information. Integrating models of emotion spread to forecast temporal trends in people’s emotions is more significant in the realm of public health.

Social influence and the emotion spread mechanism

According to social influence theory, individuals tend to follow the opinions of others and change their attitudes and behaviors to remain consistent with the group and to meet group norms (Li et al. 2018). Emotional contagion, as a dynamic transition process in crowds, can be triggered by the emotional state of others during human interactions. Event reporting combined with emotional contagion contributes to a basic form of public opinion. The study of the emotional spreading mechanism is highly important for expanding the theoretical horizon of public opinion.

As illustrated by emotional contagion theory, emotions are transmitted as information from the initiator to the receiver. The imitation–feedback mechanism is used to explain the mechanisms of emotion spread (Zhang Qiyong and Jiamei, 2013). People feel emotions from information transmitted by others and produce perceivable emotions to exhibit their emotional experience (Falkenberg et al. 2008). Furthermore, people tend to automatically and continuously mimic and synchronize others’ movements and facial expressions. Another mechanism of emotional contagion called category activation does not depend on mimicking bodily expression. This type of mechanism has proven to be important in the contagion of nonverbal emotional expressions.

On social media, emotional spread occurs alongside the dissemination of emotional information. Netizens tend to spontaneously categorize the text of posts as reflecting a particular emotional state, and the activation of emotion can trigger the associated emotional state among netizens (Peters and Kashima, 2015). After perceiving the information and emotional state transmitted by others, netizens form their own emotional state after personal processing and evaluation (Chiang et al. 2021); thus, sentiment can mutate. This means that on platforms such as Weibo, bloggers post information imbued with emotional states. Users read this information, process it, and then express their views and emotions. One study revealed that posters’ emotional expressions can determine the number of comments and emotional expressions in comments (Zhang et al. 2021). Different emotions may elicit varied responses from the audience. Netizens post comments based on their emotions after they receive information.

Regarding the dynamic mechanisms of emotion spread, from a group statistical perspective, emotions are influenced by neighbors (van Haeringen et al. 2022). The dyadic-relations-based mechanism defines emotional contagion as the continuous emotional exchange between two individuals, influenced by their connections and attributes. Benefiting from the concept of collective prevention, most studies employ epidemiological models to simulate the spread of emotions within populations. Based on epidemiological mechanisms, spreading is seen as a continuous process, with emotional contagion occurring in individuals who are carriers of emotional infection, while individuals in susceptible and recovery states cannot transmit emotions to others. Additionally, the emotional spread is influenced by factors such as emotional valence, the contagion intensity of different emotions, paths of emotional contagion, and individuals’ susceptibility (Fan et al. 2016; Wei-dong et al. 2015; Xu et al. 2022; Ferrara and Yang, 2015a). Emotional propagation on social media following natural disaster events can have dramatic individual and societal consequences. Therefore, considering the factors influencing emotional spread processes, constructing dynamic models from a population perspective to predict the process of emotional spread, and evaluating the effects of emotional spread interventions could contribute to the formulation of public sentiment guidance measures and disaster recovery efforts.

Methods of modeling the spread of emotion on social media

Currently, methods for modeling the spread of emotions on social media include data-driven modeling (e.g., time series models and neural networks) and mechanism-driven modeling (e.g., transmission dynamics models and agent models). Data-driven modeling focuses on the linear and nonlinear relationships between a model and data and relies heavily on the authenticity and reliability of the data. For example, one study developed a data-driven understanding of public attitudes toward natural disasters by investigating public sentiment characteristics via machine learning techniques (Dong et al. 2021). The sentiments in 40,000 tweets related to natural disaster topics were classified using eight machine learning models, which could provide new insights and methods for disaster management. However, the study did not consider the impact of other factors, including the process of emotion generation and stage transition, on the performance of the model. Some studies elucidate the interaction in emotional communication through social network analysis. They discuss the impacts of emotional intensity, cascade depth, and user influence on emotion spread (Miller et al. 2011). By employing a large-scale hyperlinked network with nearly 8 million nodes, this study revealed the patterns of emotion spread in the network. However, the study’s findings may be limited to specific contexts, and their generalizability to other fields may be constrained. The mechanism-driven model is based on the emotional transmission process, which refers to the transmission process of infectious disease. Compared with the data-driven model, it can explain the emotional transmission process, quantify emotional transmissibility, and simulate intervention measures. Given that emotional exchange is a dynamic process, some studies employ mathematical and computational modeling approaches to analyze emotion transmission mechanisms and simulate and predict dynamic processes (van Haeringen et al. 2022; Yin et al. 2021). Some scholars have developed mechanism-driven models for emotional transmission using thermodynamic theory (Bosse et al. 2009), agent theory (Bu and Wang, 2013), system dynamics (Ye Qiongyuan et al. 2017; Yang and Xie, 2020) and infectious disease models. Mechanism-driven emotion modeling methods consider various factors that impact the performance of the model, thus enhancing its generalizability.

Mechanism-driven emotion spread modeling

Previous studies have shown that the emotional characteristics of a group and a virus from node to node are very similar in emergencies (Durupinar et al. 2016; Durupınar, 2010). Most studies on emotion transmission utilize the infectious disease model, including crowd emotion transmission (Wang et al. 2016), group negative emotional transmission (Yin et al. 2022; Shi et al. 2021), and the transmission of positive and negative emotional states in social networks (Fowler and Christakis, 2008). While infectious disease models arise from the mechanisms of transmission, they are more capable of simulating the processes of emotion generation, variation, and evolution than traditional statistical models. For example, several researchers have used the traditional susceptible–infected–recovered (SIR) model to simulate the effects of emotional contagion on the dynamic aggregation process of virtual pedestrians (Nan et al. 2017). They successfully combined the social force model with the SIR model to capture the complex interactive relationship between population dynamic aggregation and emotion transmission. However, their method is more suitable for simulations in small-scale scenarios and requires additional CPU and GPU power for larger-scale population simulations. The susceptible–infected–recovered–susceptible (SIRS) model further suggests that the process of emotional contagion is also affected by emotional recurrence (Liu et al. 2021). The simulation was conducted using the SIRS model on a small-world network, presenting a novel approach to secondary emotion transmission in virtual space. However, the lack of real-world data sources has hindered the validation of the model’s reliability. As emotions spread on social media, netizens process and evaluate information and then produce an emotional response. However, the above models did not consider the differences in transmission patterns of different emotions and users’ emotional choices.

To improve the effectiveness of the model, several studies have examined the rules of emotion transmission under the refinement of emotional valence by considering the decisive role of users and their emotion choices. Wang et al. developed an emotion-based susceptible–infected–recovered (E–SIR) information propagation model by dividing netizens’ emotions into five categories: no emotion, anger, joy, disgust and sadness (Wang et al. 2019). They described the infection abilities of different emotions and the possibility of information transmission among users with various emotions. By establishing an SIR model that considered different emotions and connection relationships, the study expanded the scope of traditional information propagation models through collecting real-world data for model training. However, the acquisition of key model parameters was limited to curve-fitting methods, which did not allow for the extraction of model parameters from simplified real-world scenarios. Moreover, the study did not simulate or assess the effectiveness of various prevention and control strategies. The emotion-based susceptible–forwarding–immune (E–SFI) propagation dynamic model was further developed based on the quantity of forwarded Weibo messages to understand the impact of forwarding users’ emotional choices on information propagation and public sentiment formation (Yin et al. 2021). The model’s reliability was enhanced by combining actual data with sensitivity analysis, and the study compared and analyzed the effects of different emotions on information propagation. However, the study did not delve into the rationality and real-world significance of parameter values in the model. The dynamic multiple negative emotional susceptible–forwarding immune (MNE–SFI) model has been used to examine how the four sentiments of fear, shock, sadness, and anger spread on social media and the impact of sentiment mutation (Yin et al. 2022). Building upon the E–SFI model, this model further elucidated the complex mechanisms of emotion spread, considering users’ emotional choices and the possibility of emotional variation. However, the parameter estimation was solely based on curve fitting and lacked a rigorous statistical analysis of the parameters. Additionally, the model lacked sensitivity analysis and stability testing. The emotional infection unsusceptible–susceptible–optimistic–pessimistic (SOSa–SPSa) model explained the entire emotional contagion process during public health emergencies by considering four groups of people (unsusceptible–susceptible–optimistic–pessimistic) (Ni et al. 2020). The model revealed the influence of users’ emotional tendencies on the propagation of collective emotions. However, real-world data sources for model validation are lacking.

In summary, infectious disease models have been extensively applied in modeling emotion spread on social media. Considering the impact of different emotions and their relationships on the pattern of emotion spread and validating the model with real-world data are important approaches to make emotion spread modeling more realistic. Most modeling studies have estimated parameters by fitting reported emotional data, which ignores the uncertainty of parameter values when several parameters are fitted by the model. The directly calculated parameters for the model can reduce the number of estimated parameters and improve the model’s authenticity. Based on those models, this study further calculated the proportions of three types of comments infected by three types of posts and the time interval from the post timestamp to the comment timestamp, which are two key parameters used in our emotion transmission model.

Chinese context of emotion spread on social media

Social media platforms such as Sina Weibo are widely used in China, serving as important channels for information dissemination and effective media for the transmission of public sentiment. Many studies have utilized social media data to analyze emotion spread during sudden events in China. For instance, one study on emotional information dissemination on Weibo during the COVID-19 period presented a multilayer diffusion pattern of emotional messages and followed network step flow models (Yi et al. 2022). Some researchers have also referenced or adapted epidemiological SIR models to predict the process of emotional information dissemination on social media platforms such as Sina Weibo and TikTok. In this process, a user transitions from a susceptible state to an infected state under the influence of spreaders. Over time, users may lose interest in a given topic, becoming immune to it (You et al. 2022; Shen et al. 2022). Wang et al. highlighted that emotion propagation depends on factors such as information-spreading probability and the proportion of retweets for a specific type of emotion by a user reposted from another user (Wang et al. 2015). Other researchers have discovered that emotion spread models built considering factors such as spreading probability, the possibility of emotion mutation, and emotional transforming weights perform better using Sina Weibo data (Wang et al. 2017). Therefore, building upon findings from previous studies, this study categorizes users into three states: susceptible, emotionally infected, and recovered. Furthermore, the emotion spread model in this study considers factors such as the transmission rate, different proportions of emotion spread, time intervals for different emotions, and emotion decay. Unlike the typical parameter estimation methods in previous studies, which often rely on curve-fitting and manual parameter-setting, this study also employs statistical descriptions to obtain parameters, enhancing the stability and generalizability of the model.

Analysis of public emotion toward flood disasters based on social media data

Floods have become the most prevalent natural disaster worldwide. Understanding the sentiments and emotional contagion of netizens after floods can help reduce negative mass incidents and contribute to disaster prevention and mitigation. Some studies have utilized social media data to conduct emotional analyses during flood events. For instance, Dhivya Karmegam and Bagavandas Mappillairaju (2020) employed GIS analysis to understand the spatiotemporal distribution of negative emotions on Twitter during floods in Chennai, India (Karmegam and Mappillairaju, 2020). Additionally, scholars conducted a series of negative binomial regressions to examine the effects of both linguistic style and content factors on the virality of disaster messages using tweets from the 2013 Colorado floods (Lee and Yu, 2020). In the aftermath of flood events in southern China, Ma et al. (2023) utilized correlation analysis to examine the association between floods and public sentiment (Ma et al. 2023). Furthermore, scholars from China analyzed the spatiotemporal characteristics of public emotions during an extreme flood event in Henan Province on 20 July 2021, identifying differences in public emotional tendencies during different phases of the disaster (Wang et al. 2024). While those studies could provide empirical evidence for formulating public sentiment management measures following flood events, modeling approaches that are independent of data have greater predictive and alerting significance. However, both within and outside the Chinese context, relatively few studies have simulated and predicted the evolution of public emotions during flood events based on emotion spread mechanisms.

The government plays an important role in spreading positive emotional orientations of messages (Yi et al. 2022). Understanding how intervention measures impact emotion spread can help in designing more effective disaster response strategies. However, there is almost no research simulating social media intervention measures in models and analyzing the impact of interventions on guiding public emotion after flood disasters. This study simulated intervention measures such as intercepting transmission paths and changing the magnitude of positive and negative emotions. By understanding which intervention measures are more successful in controlling negative or promoting positive emotions, authorities can adjust their responses to mitigate the emotional impact of disasters on netizens.

Research questions

This study aims to explore the spread of emotions on social media after a natural disaster and propose intervention strategies to guide public opinion by addressing the following research questions:

RQ1: What are the temporal trends of three emotional orientations (positive, neutral, and negative) of posts and comments on social media after a flood disaster occurs (description)?

RQ2: How does emotion spread from posts with three different emotional orientations (positive, neutral, and negative) to comments with these three types of emotional expressions on social media when a flood disaster occurs (mechanism)?

RQ3: What are the appropriate measures for creating a better online community after a flood disaster occurs (application)?

Methods

Study design

This study is divided into five sections (Fig. 1). First, data were crawled from Sina Weibo with a topic tag for natural disasters, and we categorized the emotions in the posts and comments. Second, we assumed that netizens were influenced by the information posted and comments on posts based on their emotions. The emotions of commenters may or may not have been the same as those of the posters. We developed an emotion-based post–susceptible–comment–removed (PSCR) model according to the emotion transmission mechanism. Third, we obtained the parameters used in the PSCR model through firsthand data analysis, curve-fitting and assumptions. Fourth, we proposed a comprehensive indicator to quantify transmissibility. Finally, we conducted three intervention scenarios to identify appropriate measures to guide public opinion.

Fig. 1: Study design and model framework.

In the study has five processes, including data collection, parameters estimation, model development, transmission quantifying and Intervention evaluation.

Selection of natural disaster events for investigation

Climate change can easily lead to natural disasters such as flash floods. In recent years, people have been interested in traveling to wild and undeveloped scenic areas, as recommended by some social applications, where they may encounter threats to their lives due to natural disasters. A recent safety incident in a wild scenic hot spot was the 8·13 flash flood in the Longcaogou Scenic Area, Sichuan Province, China, which left 7 people dead and 8 injured. The flood aroused widespread discussion on social media, especially regarding the attribution of responsibility. Longcaogou has been widely recommended on social media platforms as a “new Instagram-worthy location” and a “niche holiday resort”. The Longcaogou River managers placed safety warning signs around the area, and relevant staff with horns warned tourists that the area was dangerous and that entry was not allowed. However, some tourists still chose to enter the water despite the signs and attempts at dissuasion. This event stimulated public discussion about emerging environmental health threats in some wild Instagram-worthy locations; children injured in the disaster; people’s low risk perception; and the asymmetry among government regulations, health communication and public health literacy. The 8·13 flash flood in the Longcaogou Scenic Area stimulated public opinions and emotional responses because it combined the general characteristics of a public emergency and specific features. In particular, a video of a father and son trapped in the flash flood sparked heated debate online. Thus, this study collected microblogging data related to the topic #Pengzhou flash floods#.

Data collection and sentiment analysis

On the Chinese Sina microblog, a topic can trigger discussion among online users. Netizens can create original posts that include a topic tag related to the event. Users can also express their own views and post comments under the original post. We used Python (Python Software Foundation) to crawl real records under the topic tag #Pengzhou flash floods# on the Weibo web platform from August 13, 2021, to August 20, 2021. The real records included the Weibo username, user ID, posting time, blog post content, first-level commenter username, first-level commenter ID, first-level comment time, and first-level comment content. We then cleaned the data using Python; in this process, we removed all stop words, typos, garbled characters, special characters, and duplicate information from the text through noise removal. Ultimately, a total of 556 posts and 12447 comments were included in our study. We used GooSeeker’s word separation text analysis platform to separate the full text information into words and statistical word frequency and conduct keyword extraction. Sentiment analysis was performed on the extracted keywords using GooSeeker’s sentiment analysis function (Tao et al. 2019).

Model development

Considering the process of emotion transmission and key factors that affect it, we developed an emotion-based post–susceptible–comment–removed model, which was based on the previous COVID-19 model that indicated the process by which local cases were infected by imported cases (Zhao et al. 2020). The equation of the PSCR model is as follows:

$$\frac{{dS}}{{dt}}={-\lambda }_{i}S=\,-k* \frac{{P}_{t-j}}{N}b\mathop{\sum }\limits_{j=-1}^{1}\frac{{q}_{{ji}}}{{\sigma }_{{ji}}}S$$

$$\frac{d{C}_{i}}{{dt}}={\lambda }_{i}S-{\gamma C}_{i}$$

$$\frac{d{R}_{i}}{{dt}}={\gamma C}_{i}$$

where * denotes the convolution, k denotes the convolution kernel, P_t-j denotes a function of post waves that change with time, and i and j represent −1, 0, and 1.

In the equation of the PSCR model, the total user population (N) represents all users of Sina Weibo who might have been involved in this natural disaster (including post publishers, commenters, and people who only read posts and comments). They were divided into ten categories of users, including susceptible users (S) without any emotion, commenters with three types of emotions (negative, neutral and positive orientation) (C_i), three types of removed users (R_i) and three types of posters (P_j). Once posters (P_j) post blogs, susceptible users (S) may read and post comments and sequentially transform into users in comment (C_i) states. The force of infection is defined as λ, which can be affected by the quantity of users in P_j states. The impact exists once a microblog is posted, but it gradually weakens. Therefore, we introduce the convolution kernel (k) to simulate the weakening influence of the posts. The three types of P have different influences on S; that is, a user is influenced by an emotional posting and subsequently posts a comment with his or her own emotion, which may or may not be the same as that of the influencer. We evaluate the transmission risk of S from different types of P, which is caused by two main factors: the time elapsed from the j type of post timestamp to the i type of comment timestamp (σ_ji) and the proportion of the i type of comment infected by the j type of post (q_ji). Notably, not all S are transferred into C after reading posts, with a transmission rate defined as b in the model. After users move from the S state to the C state, they may retain the emotion for a period and then move to the R state. The retention times of various types of emotions differ. We define the emotion retention time as 1/γ. Thus, γ_i is defined as the removal rate of type i comments.

Obtaining parameters affected by emotion transmission

Many factors, such as the number of emotional posts, removal rate of emotional comments (γ), and force of infection of the post (λ), can affect emotional transmission. As different types of emotions have different parameter values, we use the subscripts −1, 0 and 1 to define negative, neutral and positive emotions, respectively. This study adopts firsthand data analysis, curve-fitting, and assumptions to obtain the parameters used in the emotion transmission model.

Two key model parameters are calculated based on firsthand data. The first is the proportion of type i comments infected by type j posts (q_ji). We employ the bootstrap method to estimate the possible values of the parameter q_ji. For each iteration, we randomly select 400 of 556 blog posts to analyze the proportions of comments with the three different emotions (positive, neutral, and negative orientation) among the various emotional posts. We repeat this process 1000 times and calculate the 25^th, 50^th, and 75^th percentiles of the results of q_ji. The equation of q_ji in each iteration is as follows:

$${q}_{{ji}}=\frac{{Cumulative\; number\; of}{j\; type\; of\; posts}}{{Cumulative\; number\; of}{i\;type\; of\; comments}}$$

The results show that the median proportions of neutral comments under negative posts (q_-10 = 0.44 [interquartile range: 0.34–0.56]), neutral posts (q₀₀ = 0.49 [interquartile range: 0.39–0.58]), and positive posts (q₁₀ = 0.57 [interquartile range: 0.26–0.70]) are greater than the proportions of positive and negative comments under the three types of emotional posts (Fig. 2). The proportions of negative comments are from negative posts (q_-1-1 = 0.39 [interquartile range: 0.25–0.49]) and neutral posts (q_0-1 = 0.32 [0.21–0.39]) more than from positive posts (q_1-1 = 0.22 [0.09–0.38]). The median proportions of positive comments from negative posts (q_-11), neutral posts (q₀₁) and positive posts (q₁₁) are 0.17 [interquartile range: 0.11–0.24], 0.20 [0.13–0.25], and 0.22 [0.08–0.39], respectively.

Fig. 2: Analyzing proportion of comments with three different emotions (positive, neutral, and negative) infected by posts with three different emotions (q_ji).

A represents the first quartile of the proportion of comments with different emotions under the posts with different emotions after resampling 1000 times from the original data by Bootstrap. B represents the median percentage of the proportion of comments with different emotions under the posts with different emotions after resampling 1000 times from the original data by Bootstrap. C represents the third quartile of the proportion of comments with different emotions under the posts with different emotions after resampling 1000 times from the original data by Bootstrap.

Another parameter calculated from the data is the time elapsed from the j type of post timestamp to the i type of comment timestamp (σ_ji). We calculate the difference between the date of each blog posted and the date of each comment posted. Furthermore, we calculate the 25%, 50% and 75% quantiles of parameter σ_ji. The median time elapsed from the negative post timestamp to the negative comment timestamp (σ_-1-1 = 8.24 [interquartile range: 3.13–17.23]), the neutral comment timestamp (σ_-10 = 7.35 [interquartile range: 2.38–16.68]), and the positive comment timestamp (σ_-11 = 7.04 [interquartile range: 2.53–15.95]) is greater than the time elapsed from the neutral and positive post timestamps (Fig. 3). All of the values of parameter σ_ji fit an exponential distribution (Supplementary Fig. S1).

Fig. 3: Analyzing time elapsed from the timestamp of post to the timestamp of comment (σ_ji).

A–C show the cumulative probability density function of σ_ji matrix. The intersections of the three dashed lines parallel to the x-axis with the three colored curves in A–C are plotted in D–F, respectively. D, E, and F represent the 25%, 50%, and 75% percentile of σ_ji.

Previous studies on emotion-based models for sentiment-spreading set the number of initial users at approximately 200000. With the APIs of Sina Weibo, one study set 219837 users as the original users to spread the information (Wang et al. 2017). Some scholars estimate that the size of the initial susceptible population for different events on Sina Weibo ranges from 2.64 × 10⁴ to 2.33 × 10⁵ through data-fitting (Yin et al. 2021; Yin et al. 2022). None of these studies can ensure the accuracy of the total number of users due to the inconsistency in the number of internet users involved in each event. Thus, this study assumes that the susceptible population is 200000 users and sets the total population (N) at 200000.

We obtain three parameters (transmission rate from post to comment [b], removal rate of comment [γ_i] and initial value of compartment C [C_i-0]) through curve-fitting since we cannot find them in the data analysis or references. These parameters are detailed in Supplementary Table S1.

Transmission quantification

We cannot define the basic reproduction number (R₀) of the PSCR model since it is a secondary infection model. Therefore, we introduce an indicator named the secondary attack rate (SAR: defined as the probability that an infection will occur among susceptible people within a specific group [e.g., household or close contacts]) sourced from infectious disease to quantify emotional transmissibility. In this study, the SAR is defined as the probability that an emotional poster will experience emotional comments among susceptible users. The equation of the SAR model is as follows:

$${{SAR}}_{{ji}}=b\frac{{q}_{{ji}}}{{\sigma }_{{ji}}}{/\gamma }_{i}$$

Intervention simulation

In this study, we simulate three intervention scenarios (Supplementary Table S2), including I) interrupting the transmission route, II) changing the number of emotional posts, and III) changing the numbers of positive and negative posts. All the intervention effects are evaluated by the cumulative numbers of negative comments, positive comments or neutral comments. We simulate the variation of the three scenarios by introducing a parameter of scale factor (x). For example, we simulate the effect of interrupting each transmission route (Scenario I) by gradually decreasing x. We simulate the changing number of negative posts (P_-1) and positive posts (P₁) separately in Scenario II. In Scenario III, we simultaneously change the number of P_-1 and P₁ and further calculate the changes in the three indicators.

Statistical analysis

The data are captured by Request in Python 3.9.12. The PSCR model is performed with Anylogic 8.7.0 (Personal Learning Edition), and the sum of squares of deviations (RMSE) is used to judge the goodness of fit. All the plots are drawn with the Seaborn and Matplotlib packages of Python.

Results

Temporal trend of three different emotion-oriented posts and comments on natural disasters

Figure 4 shows the temporal trends of posts and comments with three different emotion orientations (positive, neutral, and negative emotions). The overall sentiment of posters was negative after this flood, and there were three epidemic peaks. The sentiment of comments was mainly neutral (C₀ = 5786), followed by negative emotions (C_-1 = 4063). There were also three epidemic peaks in the curves of the emotional comments. The first comment peak lasted approximately 6 hours and generated a total of 2752 comments. The second comment peak lasted approximately 24 hours and generated a total of 5802 comments. The third comment peak lasted approximately 24 hours and generated a total of 2892 comments. The times of the three epidemic peaks of the emotional comments were similar to the peak times of the posts.

Fig. 4: Curves for the number of post and comment per hour for three types emotion.

Plot A shows curve of three types emotion of post. Plot B shows curve of three types emotion of comment.

There were more negative posts (P_-1C₀ = 3541, P_-1C_-1 = 3082, P_-1C₁ = 1318) than positive posts (P₁C₀ = 1662, P₁C_-1 = 591, P₁C₁ = 613) and neutral posts (P₀C₀ = 390, P₀C_-1 = 583, P₀C₁ = 241). These results answer RQ1 and indicate that posts with negative emotions tend to trigger more intense discussion among users than posts in the other emotional categories (neutral and positive orientation).

Model validity and transmissibility

The PSCR model we constructed can generally fit three types of real-time comment curves (RMSE < 51) (Supplementary Fig. S2). The curve fit by the model is consistent with the actual reported curve, which indicates that our PSCR model can be used to simulate the process of emotion transmission on social media for the 8·13 flash flood in the Longcaogou Scenic Area.

A large value of SAR_ji represents high transmissibility from posts to comments. Our results show that the transmissibility of the three types of comments caused by neutral posts (SAR_0-1 = 12.17, SAR₀₀ = 18.86, SAR₀₁ = 7.46) is greater than that of comments caused by negative posts (SAR_-1-1 = 4.77, SAR_-10 = 6.03, SAR_-11 = 2.43) and positive posts (SAR_1-1 = 1.39, SAR₁₀ = 5.85, SAR_1-1 = 1.79) (Fig. 5A–C). The SAR_ji results show that the emotions in comments are affected by posts with various emotional orientations, and the abundance of negative emotional comments is mainly caused by neutral and negative post transmission. The PSCR model and the value of SAR_ji can explain the mechanism by which emotion spreads from posts with the three different emotional orientations (positive, neutral, and negative) to the three different types of emotional expression comments on social media (RQ2).

Fig. 5: Secondary attack rate (SAR) of each transmission route from post to comment.

Plot A–C represent 25%, 50% and 75% quantiles of SAR, respectively.

Intervention simulation

This study conducted three intervention scenarios to analyze the appropriate measures to create a better online community after a flood disaster occurs (RQ3). These scenarios included interrupting the transmission route, changing the number of emotional posts, and changing the number of positive and negative emotion-oriented posts.

Transmission interruption

Effective interventions to interrupt transmission are essential to protect susceptible users and cut off information transmission paths. To reduce the number of negative comments, fully interrupting the transmission route from negative posts (reducing the number of negative comments by 30.04%) has a greater impact than fully interrupting transmission from positive posts (reducing the number of negative comments by 15.42%) (Supplementary Fig. S3). To reduce the number of neutral comments, interrupting the transmission route from neutral posts (reducing the number of neutral comments by 39.37%) has a greater impact than interrupting transmission from negative posts (reducing the number of neutral comments by 25.67%) and positive posts (15.72%) (Supplementary Fig. S4). Fully interrupting the transmission route from positive posts can reduce the number of positive comments by 18.87%, and fully interrupting transmission from negative posts can reduce the number of positive comments by 14.53% (Supplementary Fig. S5). These results indicate that the primary transmission source for any emotional comment is posts with the same emotion. To reduce public expressions of negative emotions on social media, it is necessary to focus on guiding bloggers to reduce their negative posts.

Changing the number of emotional posts

Figures 6, 7 show the effect of reducing the number of posts with positive and negative emotions by 10% for the number of comments with three different emotions on social media. As shown in Fig. 6, reducing the number of negative posts from 100% to 50% can effectively reduce the quantity of negative comments by 14.97% and reduce the quantity of positive comments by 7.17%. Reducing the proportion of negative posts may reduce the quantity of positive comments but can provide better control over the quantity of negative comments. As shown in Fig. 7, when the number of positive posts increases from 100% to 150%, the number of positive comments increases by 9.40% and the number of negative comments increases by 7.64%. Increasing the number of positive posts may increase the quantity of negative comments, while the intervention of increasing positive posts can provide a more positive online atmosphere.

Fig. 6: Simulating the effect of changing on negative post (P₋₁) numbers for emotional comment numbers.

P and C represent posts user and comment users. Subscript −1, 0 and 1 represent the negative, neutral and positive emotion, respectively. A, C and E indicate the new number of emotional comments per hour, and B, D and F indicate the cumulative number of emotional comments.

Fig. 7: Simulating the effect of changing on positive post (P₁) numbers for emotional comment numbers.

P and C represent posts user and comment users. Subscript −1, 0 and 1 represent the negative, neutral and positive emotion, respectively. A, C and E indicate the new number of emotional comments per hour, and B, D and F indicate the cumulative number of emotional comments.

Simulation of interventions that change the number of positive and negative emotion-oriented posts

This study further evaluates the effect of combining changes to the number of positive and negative emotion-oriented posts on the quantity of comments with negative emotions (Fig. 8A). Reducing the number of positive and negative emotion-oriented posts can reduce the quantities of comments with negative emotion, although negative emotion-oriented posts have a large influence. For example, to reduce the number of negative comments, adjusting 40% of positive posts is equivalent to adjusting 20% of negative posts. Figure 8B shows the intervention effect of changing the number of negative and positive posts on increasing the quantities of positive comments. The influence of changing the proportion of positive posts is large; for example, adjusting it by 27% is equivalent to adjusting the proportion of negative posts by 21% to enhance the quantity of positive comments. Reducing the number of negative posts is more effective than increasing the number of positive posts for reducing the relative ratio of negative to positive posts (Fig. 8C). Simultaneously, increasing the number of positive posts and reducing that of negative posts helps create a better online community. These results indicate that decision-makers can control the process of emotion spread based on the characteristics of the event.

Fig. 8: Combined effects of changing negative post (P₋₁) and positive post (P₁) numbers on negative comment (C₋₁) and positive comment (C₁) numbers.

A and B represents the cumulative number of positive comments and cumulative number of negative comments, respectively. C is ratio of plot A to plot B.

Sensitivity analysis

A sensitivity analysis of parameters is adopted because the value of the total number of users (N) is assumed in this study. The sensitivity analysis indicates that there was no major change in the three emotional curves even if the value of N increased from 20000 to 240000 (Supplementary Fig. S6). Therefore, it can be suggested that parameter N is not sensitive to the model, which also indicates that it is reasonable for this study to set N = 200000 in the emotion transmission simulation.

Discussion

Based on the analysis of emotional trends and descriptive statistics, this study proposed a PSCR model to understand the emotion transmission of posts with three different emotional orientations to comments with three different types of emotional expression on social media when a flood disaster occurred. We proposed a new idea for modeling emotion transmission on social media that applied the parameters of the proportions of comments with three different emotions infected by different emotional posts (q_ij) and the elapsed time from the post timestamp to the comment timestamp (σ_ij) to the transmission dynamic model. Our results show that the number of emotional posts plays an essential role in mitigating harmful emotion transmission and promoting meaningful emotion diffusion.

Changes in netizens’ emotions during flood events

Scholars have indicated significant associations between the emotional expressions of posts on social media across time and the context of disasters (Gruebner et al. 2018). The occurrence of emergencies is accompanied by changes in public sentiment. In this study, the peak was concentrated during daytime hours within three days of the flood disaster event. The curves of emotional blog posts and comment quality were consistent with the actual situation. The overall sentiment of commenters after the flood was neutral, and there were more negative than positive opinions. After the 2020 flood in the middle and lower reaches of the Yangtze River Basin in China, a sentiment analysis also revealed that netizens’ emotions were mainly neutral (Guo et al. 2021). Some negative emotions expressed were generated by fear and anxiety about the flood (Karmegam and Mappillairaju, 2020). Unlike other flood events, the 8·13 Pengzhou flood caused some netizens to use anger-related words after they received information about human casualties. Some netizens blamed the injured themselves and platforms that promoted tourist attractions, such as the Little Red Book. We suggest that the government, social media managers and tourist attraction recommendation platforms cooperate with health educators to reduce preventable casualties in natural disasters and their emotional impact on netizens.

Negative emotions are more likely to spread on social media and persist longer

Emotional microblogs with emotional words may elicit physiological arousal from users and lead to more discussion, especially those with negative emotions (Stieglitz and Dang-Xuan, 2013). In the expression of public opinion, “negative preference” often obtains more social capital. Netizens tend to vent negative emotions to release stress and ease their mood (Fan et al. 2019). This study revealed that negative emotional comments are caused mainly by neutral and negative post transmission, and the average elapsed time from a negative post timestamp to a comment timestamp is longer than the elapsed time from a positive post timestamp or neutral post timestamp. In other words, posts with negative emotions are more persistent and contagious than are posts with positive emotions. Furthermore, negative posts evoke more public attention and discussion than positive sentiments (Zeng and Zhu, 2019). Increasing the number of negative emotion posts increases the duration of emotional contagion (Chen et al. 2018). One study conducted a questionnaire survey and found that emotional contagion impacts netizens’ negative emotional communication (Lu and Hong, 2022). Negative emotions on social media can easily infect others, and individuals can quickly turn into spreaders of negative information. Thus, in terms of the large influence of negative information on social media, the findings of our study are consistent with those of other studies. This study introduced a comprehensive indicator (SAR) to quantify the transmissibility of emotional information. Some studies have also proposed using R₀ to quantify transmissibility (Shen et al. 2022; Yin et al. 2022). R₀ is indeed better at assessing emotional transmission because its definition is easier to understand and accept. However, those studies were limited to assessing the transmissibility of different transmission routes. The SAR proposed in this study can be effectively applied to the PSCR model and used to compare the transmissibility of different transmission routes. This study provides a new method for quantifying the transmissibility of online emotion information and a key parameter for modeling emotion transmission with various categories of emotions. Our proposed SAR can be applied to the PSCR model, and the transmission abilities of different transmission routes can be further compared. This approach provides not only a new method for quantifying communication ability but also effective parameters for different types of emotional communication models. The findings could be valuable for the provision of emergency management. We suggest that emergency managers monitor and attempt to reduce the number of negative blog posts.

PSCR model simulates the emotional transmission process in the real world

Some users immediately copy the emotions generated by original posts, and some users may post content with the same or different emotions as an original post after processing information. Considering users’ emotional choices, the elapsed time from the timestamp of posts to the timestamp of comments, and the proportions of three types of comments infected by three types of posts, this study developed a PSCR dynamic model of multiple temporal information in messages. The PSCR model improves upon the traditional emotion-based SIR model. Unlike the traditional SIR model and the SIRS model, the PSCR model considers how posts with different emotions (i.e., positive, negative and neutral) impact commenters and considers a situation in which commenters express their emotions after processing information. Other studies also support the consideration of the infectiousness of various emotions and the possibility of information transmission between different emotional users. For example, the E–SIR model suggests that susceptible users can be influenced by forwarding users who are exposed to information with three possible emotional states: positive, neutral, and negative (Yin et al. 2021). The E–SFI model considers how different emotions impact the spread of information and the possibility of information transmission between users by dividing users’ emotions into five categories: no emotion, anger, joy, disgust and sadness (Wang et al. 2019). The SOSa–SPSa model explains the entire contagion process by considering four groups of people (unsusceptible–susceptible–optimistic–pessimistic) (Ni et al. 2020). The MNEs–SFI model discusses information transmission with four negative emotions. Most models estimate the parameters needed to study the mechanism of emotion transmission and simulation by fitting reported emotional data. Most parameters obtained through model fitting increase the uncertainty of parameter estimation, which leads directly to the model’s inability to fit the actual situation. Our model calculates the time interval from posters to commenters (1/σ) and the proportion of three emotional comments associated with different emotional posts (q_ji) from real-world data with a few parameters obtained via model fitting. That is, this study proposes a new idea for simulating emotion transmission with a dynamic model after statistical analysis of the parameters. This new idea can minimize uncertainty in model fitting. In addition, the influence of posts weakens over time. This study innovatively introduces the kernel of convolution (k) to simulate the weakening influence of posts in the PSCR model. The RMSE results indicate that the PSCR model can simulate the emotion transmission process in the real world. Our findings are consistent with previous literature (Yin et al. 2022; Yin et al. 2021), suggesting that the dynamic model is suitable for simulating emotion transmission. This study provides theoretical implications for the literature on the simulation of emotion transmission on social media during sudden public events by applying a new modeling approach. Other studies can further verify the suitability of this approach for simulating emotion propagation in other public emergencies.

Regulating emotional transmission paths and numbers can help create a positive online environment

Emotional posts cause emotional contagion on social media after an emergency. Users’ emotional tendencies have significant positive effects on negative emotional communication (Lu and Hong, 2022). It is important to apply emotion transmission interventions to reduce the negative effects of exposure to natural disaster events on public mental health. There is dynamic interdependence between the number of posts and the intensity of emotions expressed in message content (Qiu et al. 2020). The simulation of strategies for emotion regulation intervention in our study shows that reducing the number of negative posts can reduce the number of emotional comments, especially negative comments. Positive and negative emotions have different effects on information diffusion and the spread of emotion in online social ecosystems (Ferrara and Yang, 2015b). Social media platforms apply interventions to regulate negative emotions, which are effective in reducing negative emotions in public opinion environments. Increasing the number of positive emotions also contributes to increasing the number of positive comments, which is helpful for creating a positive online environment. To identify effective measures to improve online psychological resilience and develop a positive online environment, this study simulated simultaneous changes in the numbers of negative and positive posts. Simultaneously, decreasing the number of negative emotion posts and increasing that of positive emotion posts reduced the relative ratio of negative to positive emotional comments. Other studies have also discussed solutions for reducing netizens’ negative opinions and comments containing incivility, such as the use of a moderation strategy that deletes inappropriate content (Boberg et al. 2018), a requirement to sign in to prevent anonymous commenting (Santana, 2019), and facilitating implicit emotion regulation in online news commenting via emotion-labeling interventions (Syrjämäki et al. 2022). This study simulated interventions by considering the mechanism of emotion transmission and quantified the effect of the interventions. Our findings can help social media platforms optimize their regulatory measures for public opinion to obtain the expected relative ratio of negative to positive comments in terms of their actual needs and economic and time costs.

The internet provides a good platform for individual expression, but there are potential negative effects regarding free speech. Improving the ability to technically control the internet can somewhat limit online communication but may violate citizens’ freedom of speech. Therefore, attempts to improve online public expression could focus on balancing freedom of expression, transparency of information and social stability. A well-functioning public sphere is a necessary condition of a well-functioning democracy (Miksa, 2017). After a natural disaster, health education workers and social media managers should monitor public opinion in real time and take actions to reduce the potentially negative impact of media reporting of public emergencies on public emotion. They can collaborate with the public and journalists to guide public opinion through an open and transparent debate carried out in the public sphere with respect to people’s rights and ensure people’s privacy in the process.

Study strengths and limitations

This study applied a PSCR model to explain the transmission process of posts to comments and simulated interventions that changed the numbers of positive and negative original posts to explore strategies to guide public opinion. This study makes practical contributions for health educators and online community managers to develop optimal communication intervention strategies to reduce the continuous impacts of negative public opinion after natural disasters. The spread of emotions on social media has been studied recently; however, research is lacking on changing the numbers of posts across emotions to shape patterns of emotional contagion. We performed scenario simulations in which social media managers and opinion leaders changed information dissemination, such as encouraging bloggers to post positive content and reducing negative posts. Our findings provide theoretical implications. Our study further verified the applicability of emotional contagion theory in the area of emotion transmission on social media after flood events. The 8·13 Pengzhou flood involved preventable casualties and children, which sparked heated public debate and emotional fluctuations. In the future, researchers can consider using the modeling ideas of this study or the PSCR model to analyze emotional communication and conduct intervention simulations for other events that cause heated public discussion and involve personal and public opinions in the online community.

Our study also has some limitations in the development of the model, data collection and parameter estimation. In terms of the development of the model, this study classified emotions into three categories. Some studies have employed a finer-grained categorization of emotions, which is highly meaningful because it yields more precise results in terms of emotional valence and the calculation of key parameters for emotion propagation. Nevertheless, an excessive number of emotion categories significantly increases the complexity of models, leading to exponential growth in the number of parameters needed for emotion transmission modeling. Models strive to simplify the complexity of the real world, necessitating a trade-off between the granularity of categorization and the number of parameters to enhance the model’s reliability. In addition, the PSCR model is developed upon assumption. We assume that all susceptible users have an equal possibility of reading Sina Weibo posts. Individual variables, such as age and daily routine, which may affect reading rates, were not considered in the model. Given that our model operates at the population level, the influence of individual factors can almost be ignored, especially in simulation modeling for large-scale samples. In terms of data collection and parameter estimation, first, this study only investigated the 8·13 Pengzhou flood, which may lead to potential bias in the generalization of this study’s findings to other events. However, the 8·13 flash flood in the Longcaogou Scenic Area exhibited characteristics commonly found in other public emergencies, such as abruptness, public impact, uncertainty, and situation imbalance. It also had special features, such as violations of government regulations, avoidable fatalities, and injuries to children, which sparked extensive public discussion and emotional fluctuations. These aspects increased the complexity of the emotional transmission process. In clarifying this complex mechanism, the transmission dynamic model offers the advantages of real-time estimation of the population’s emotional state, quantification of transmission ability, and simulation of intervention measures. Furthermore, this study initially calculated parameters related to emotional propagation, such as the proportion of comments influenced by different emotional posts and the time intervals between posts and comments. Compared with all the parameters sourced from model estimation, this study’s modeling process can minimize uncertainty in model-fitting. Consequently, the PSCR model proposed in this study, along with its modeling approach, can be applied to other public emergencies that incite fervent public discussion and emotional fluctuations. Second, although we obtained some key modeling parameters from the real world, this study calculated 4 parameters by curve-fitting, which led to nonunique model solutions. In the future, we hope to obtain more parameters and refine our modeling approach by conducting surveys. Third, the content of comments is created by first-level commenters, and netizens can comment and express their opinion under first-level comments. This study analyzed emotion transmission only from posts to first-level comments. Emotion transmission between commenters was ignored. Finally, because social media followers (fans) of social media blogs are mainly like-minded people, it is easy to generate group pressure and normative social influence, which can increase the likelihood of emotional copying. Future studies on emotional contagion on social media should calculate the weight of communication by considering the number of fans of bloggers and the interaction between fans and bloggers to make the model better fit the real world.

Conclusion

In this study, we demonstrated the trend of the emotional expressions of posts and comments after a flood disaster. Based on the analysis of the parameters of the proportions of comments with three different emotions infected by emotional posts (q_ji) and the elapsed time between the post timestamp and the comment timestamp (σ_ji), we established a PSCR model to explain the dynamics of emotion transmission from posts to comments. The proposed model can provide an accurate description of the entire process of emotional contagion in the internet environment after a natural disaster. The emotional choices of original posters play critical roles in influencing emotional contagion in public opinion. Reducing the number of negative posts can reduce the quantity of comments, especially negative comments. Increasing the number of positive posts leads to more comments, especially positive comments. Furthermore, the spread of harmful emotional expressions can be controlled by changing the numbers of negative and positive posts simultaneously. With the increasing use of digital devices and exposure to big social media data, this study can contribute to the design of measures to address negative public opinion after natural disasters. We believe our PSCR dynamics model fills a theoretical gap and supports the development of effective communication strategies on social media after a disaster event or other events to address netizens’ emotional fluctuations.