Sage Journals: Discover world-class research

Abstract

Objectives: During the early phase of the COVID-19 outbreak, misinformation spread rapidly, hindering effective health communication and fueling xenophobic violence. The politicization of health issues, along with the manipulation by social bots and astroturfing accounts, posed significant challenges. This study aims to investigate how misinformation spreads through social media, involving malicious actors like trolls and bots, and explores emotional contagion during public health crises. Methods: Using a computational methodology that combines semantic modeling, social network analysis, bot identification, emotion analysis, and time series analysis, the study analyzed over 700,000 tweets from February to July 2020. Results: The findings reveal that inauthentic actors amplified negative emotions, particularly among news and political actors, while positive emotions were less prominent. Astroturfing accounts acted as key nodes, perpetuating negative emotional contagion. Conclusion: This study provides a framework for monitoring emotional responses in public health crises, with findings applicable beyond COVID-19 to other public health emergencies.

Keywords

health misinformation public health crisis emotional contagion astroturfing social network

Introduction

The COVID-19 pandemic witnessed an unprecedented “infodemic” of misinformation with severe real-world consequences. Dangerous health misinformation contributed to vaccine hesitancy and improper treatments,^1,2 while false claims about the virus’s origins and spread fueled xenophobic violence.³ For example, the label “Chinese virus” was spread globally, which led to antipathy and rejection of Chinese people. These emotionally charged claims incited hatred and violence, with documented attacks targeting Asians during the pandemic.⁴ In one case, a Texas man stabbed a Burmese American family, erroneously blaming them for spreading COVID-19. Separately, a Brooklyn assailant doused an Asian woman with acid, causing severe burns.

Such misinformation thrived through a storm of factors: public anxiety during crises created cognitive vulnerabilities, while bots in social media preferentially amplified emotionally charged content.⁵ People tended to believe information that aligned with their preexisting worldviews or provided simple explanations for complex situations. This phenomenon proved particularly dangerous during health emergencies, when accurate information could mean the difference between life and death.

In this environment of widespread digital misinformation, globally observed inauthentic coordination tactics, including astroturfing (where artificially manufactured online support is presented as genuine public opinion), played a significant role.^6,7 State and non-state actors have exploited these tactics to amplify divisive narratives, deliberately inciting anger, fear, or xenophobic hostility in vulnerable populations. During the pandemic, weaponized misinformation fueled real-world violence, demonstrating how engineered outrage can rapidly escalate into offline harm.

This underscores the urgent need to examine emotional contagion (i.e. the spread of emotions from one person to another⁸). A large-scale Facebook experiment showed that emotions can be transmitted online, even without nonverbal cues.⁹ As people spend more time on digital media, their emotional expressions may increasingly align with those of others.¹⁰ Research has shown that social bots play a role in stirring emotions, operating on the fringes of social systems, and targeting influential individuals with violent, inflammatory content to deepen social conflict.¹¹ Bots undermine traditional actors like media personalities and experts, amplify certain political messages,¹² and often display polarized emotions during contentious events.¹³ They can also shape topic trends and spread negative emotions during public health emergencies.¹⁴

Understanding how malicious actors manipulate emotions to drive engagement and polarization is critical for countering misinformation’s societal damage. Rather than focusing solely on the human-machine divide, it is more useful to explore a broader set of actors involved in emotional contagion, examining the interactive networks among them to better understand these relationships. Our study investigates these dynamics by analyzing the roles of key actors and the evolution of emotional contagion in misinformation during COVID-19.

This study aims to utilize computational methods to uncover the networks of misinformation spreaders and the mechanisms of emotional contagion, with a focus on assessing public reactions during health crises. On the whole, the research classifies misinformation content and its contributors, then examines emotional contagion over time using time series analysis. The research questions addressed include: Actor feature mining (RQ1): What types of actors participated in spreading China-related COVID-19 misinformation? Content feature mining (RQ2): How did emotion in China-related COVID-19 misinformation vary across actor types? Actor-network analysis (RQ3): What roles did different actors play within the emotion network (i.e. a directed, weighted network modeling emotional influence among actors)? Emotional contagion analysis (RQ4): How did emotional contagion evolve temporally among actors spreading China-related COVID-19 misinformation?

Methods

Data collection and preprocessing

Twitter serves as a critical platform for studying misinformation due to its role in real-time information dissemination during health crises. Its viral features (retweets, hashtags, mentions) enable rapid narrative spread, while its public API facilitates scalable data collection. The platform also attracts inauthentic actors (bots, trolls) due to algorithmic amplification of emotional content, making it a hotspot for disinformation. Prior research on political and health misinformation often uses Twitter, allowing direct comparison with existing literature.^15–17 Compared to closed platforms, Twitter’s public interactions simplify tracking network dynamics and emotional contagion, ensuring methodological transparency and reproducibility.

The data collection process is shown in Figure 1. This study used Twitter’s public API v1.1 to collect English-language tweets about COVID-19, predominantly from North America and Europe, starting February 17, 2020. The data remained unaffected by Twitter’s verification process removal in October 2022. To focus on China-related misinformation, the search was narrowed using English keywords from Poynter’s fact-checking database,¹⁸ refined with Part-of-Speech tagging in spaCy.¹⁹ Relevant tweets were filtered for China-related entities, resulting in 341 keyword groups (examples shown in Table 1). Then, semantic modeling with Doc2vec^20,21 and HDBSCAN identified clusters of similar tweets. Finally, the study collected 203,120 original tweets, 522,599 retweets, and 406,995 users, with topics primarily covering viral origins (48.9%), conspiracy theories (24.2%), authoritarian actions (8.7%), impact (8.4%), social conflict (4.3%), medium (3.8%), and preventive treatment (1.6%). These topics reflect the Poynter’s classification schema.

Figure 1.

Data collection and preprocessing.

Table 1.

Keyword examples for each topic.

	Topics	Keywords
1	Origin/causes	Eat, bat, soup, restaurant, China
2	Origin/causes	Animals, served, meeting, 40, thousand, people, China, virus, spread
3	Conspiracy theories	Military, bioweapon, developed, China, army
4	Conspiracy theories	Lab, Wuhan, bio, weapon
5	Conspiracy theories	Chinese, spy, Canadian, lab, Wuhan
6	Authoritarian actions	Court, approval, kill, 20000, patients
7	Impact	Kill, 112000, people, infect, 2800000, China
8	Social conflict	Virus, related, treatment, Uighur, Muslims, China
9	Medium	5g, poles, trigger, symptom
10	Preventive treatment	Vaccine, China, test, bad, human

Actor feature mining (addressing RQ1)

Knowledge graph and entity analysis

To classify different verified users, this study adopted knowledge graph and entity analysis methods. Verified users on Twitter in 2020 met specific visibility criteria. For analysis, the full name of verified users was used as the entity, and the Google Knowledge Graph API and IBM Natural Language Understanding API were leveraged to obtain relevant labels²² and classify users based on returned keywords. The APIs’ keyword outputs were cross-checked against a manually annotated subset (n = 200) of user profiles. Intercoder reliability (Cohen’s κ = 0.82) confirmed strong agreement between automated and manual labels.

Social network analysis and bot detection

To classify non-verified users, this study employed social network analysis to identify instances of “astroturfing,” where the true sponsors of a message are disguised to appear grassroots-driven. Since non-verified users can include both human and bot accounts with similar clustering behavior, new identification methods were developed. Co-tweeting and co-retweeting networks were constructed using NetworkX, and Keller’s method²³ was applied to detect groups spreading the same tweet quickly. Botometer, with over 1000 features, was used to assess bot probability, setting a conservative threshold of 0.8 to classify users as bot-like or human-like.²⁴ Accounts that have been deleted or blocked were also identified as exceptional users in this study.

Content feature mining (addressing RQ2)

Emotion analysis

To examine the emotional content of misinformation, this study employed a textual emotion analysis method based on an emotion lexicon. To calculate emotion scores (the frequency of emotion words in a tweet divided by the number of words in the tweet and normalized by the number of daily tweets), we used The NRC Emotion Lexicon created by the National Research Council Canada.²⁵ It is widely used in the international academic community and provides a categorization of eight basic emotions (anger, fear, anticipation, trust, surprise, sadness, joy, and disgust) as well as positive and negative sentiments. This provides a measure of the emotional content in misinformation, which can help to shed light on the emotional appeals that may be used to spread mis- and disinformation on social media.

Actor-network analysis (addressing RQ3)

Social network analysis

This approach was also used to explore the interactions between actors and assess their roles. Accounts of the same category were aggregated, an edge represented a retweet relationship, and the size of a node represented the in-degree value. Edge weights were defined as the count of negative words in retweeted tweets, operationalizing the directional transfer of negative sentiment between actors. This approach aligns with established sentiment analysis methods²⁶ and reflects asymmetric emotional contagion relation in retweet networks, where higher weights indicate stronger negative sentiment propagation from source to target. In this study, the software Gephi was used to visualize social networks.

Emotional contagion analysis (addressing RQ4)

Time series analysis

After applying the Augmented Dickey-Fuller (ADF) test and achieving stationarity with one differencing, indicating the avoidance of spurious regression, this study primarily employed the Granger causality test to investigate the interrelationship between multiple actors. Granger causality is a statistical hypothesis test for determining whether one-time series can predict another. It does not imply true causality but rather indicates a predictive relationship based on past data. This approach aimed to examine the relationship of emotional contagion on the China-related misinformation campaign.

To address multiplicity bias, we implemented three safeguards. First, we selected the optimal lag length by requiring agreement across four standard information criteria: AIC, BIC, FPE, and HQIC. All four consistently indicated lag 10, confirming a data-driven choice. Second, we required both statistical significance (p < 0.05) and theoretically plausible effect sizes for substantive interpretations. Third, we grounded results in established theory, such as astroturfing’s known negativity bias, to distinguish meaningful patterns from statistical noise. Sensitivity analyses with alternative lags (8–12) further supported robustness, minimizing false discoveries while maintaining transparency.

Results

Actor feature mining (addressing RQ1)

Knowledge graph and entity analysis

As shown in Table 2, the largest group of verified users were in the news category (65.73%), followed by politics (9.64%). Among secondary classifications, journalists (39.95%) and news media (24.93%) were most prominent, with political figures at 7.27%. These verified users in news and politics played a key role in misinformation networks.

Table 2.

Distribution of verified user types.

Primary classification			Secondary classification
Categories	Number	Percentage (%)	Categories	Number	Percentage (%)
News	6748	65.73	Journalists	4102	39.95
			News media	2560	24.93
			News agencies	86	0.84
Politics	990	9.64	Political figures	746	7.27
			Government organizations	186	1.81
			Party organizations	58	0.56
Business	725	7.06	Companies	416	4.05
			Businessmen	309	3.01
NGOs & activists	630	6.14	Non-governmental organizations	339	3.30
			Activists	291	2.83
Entertainment	557	5.43	Artists	458	4.46
			Entertainment news	99	0.96
Academics	437	4.2%	Specialists	349	3.40
			Academic institutes	71	0.69
			Academic publications	17	0.17
Sports	180	1.7%	Sports celebrities	111	1.08
			Sports news	56	0.55
			Sports organizations	13	0.13
Total	10,267	100.00	Total	10,267	100.00

Social network analysis and bot detection

For non-verified users, astroturfing classification by automation degree (Table 3) revealed that 7.58% of astroturfing was linked to bot-like accounts, which showed typical bot behavior such as high engagement in likes and retweets. “Exceptional users” denoted accounts confirmed by the platform as deleted or blocked for policy violations. This classification was applied only when the platform explicitly labeled the account status or access attempts returned error codes. However, most astroturfing was carried out by human-like accounts, which, despite producing fewer tweets and retweets than bots, contributed the highest overall number.

Table 3.

Classification of astroturfing accounts and their characteristic.

a	Total	b	Total	Mean
a	Total	b	Total	User-like	Like	Retweet	Tweet	Repeat-tweet	Repeat-retweet
Astroturfing	157,223	Human-like	91,750	31,132.02	6.38	1.76	0.15	0.03	0.25
		Bots-like	11,921	57,288.51	24.83	2.68	0.39	0.08	0.31
		Others	53,552	35,255.21	24.3	2.22	0.69	0.19	0.28

a = primary classification, b = secondary classification.

In addition to astroturfing accounts, some unverified users were excluded from co-tweet and co-retweet networks (Table 4), including 23,925 bots (5.89% of total users) and 64,834 exceptional users (15.95%). The remaining users, not classified as astroturfing, bots, or anomalies, were considered normal users (36.96%).

Table 4.

Final classification of actors and their characteristics.

Primary classification	Mean							Total users
Primary classification	Follower	Following	User-like	Tweet	Retweet	Repeat-tweet	Repeat-retweet	Total users
Astroturfing	2644.84	2201.35	34,519.67	0.35	1.90	0.09	0.27	157,223
Bots	7393.30	3190.75	32,019.18	0.55	0.95	0.01	0.10	23,925
Exceptional users	2607.59	1892.27	20,715.08	0.44	0.87	0.01	0.09	64,834
Normal users	2977.11	1491.07	22,423.25	0.41	0.85	0.00	0.08	150,253
Entertainment	290,550.02	8393.76	28,993.67	1.51	0.50	0.10	0.10	557
Business	170,770.05	13,761.57	15,536.02	2.27	0.33	0.16	0.05	725
News	356,661.26	3425.52	11,822.46	5.47	0.31	0.45	0.04	6748
Politics	380,347.83	4631.34	11,193.61	2.04	0.31	0.10	0.03	990
Sports	406,602.05	2885.43	13,149.27	1.18	0.31	0.04	0.06	180
Academics	98,561.35	3461.05	17,145.43	2.14	0.29	0.10	0.04	437
NGOs & activists	187,759.19	4323.59	13,304.95	3.20	0.28	0.22	0.04	630

Astroturfing accounts were highly active, contributing 59.81% of retweets and 67.23% of repeat retweets. Non-astroturfing bots engaged heavily in likes (M = 32,019.18) but did not significantly spread misinformation via retweets. Influential users in news and politics had higher averages in likes and tweets, with news (M = 5.47) and NGOs & activists (M = 3.20) being particularly active in original tweets. This indicates that authentic, influential users in news played a key role in spreading misinformation, challenging the idea that automated technologies were the main drivers.

In summary, for RQ1, astroturfing accounts and bots spread misinformation mainly through retweets and likes, while authentic users, particularly in the news, played a unique role by posting original tweets. Observing the interactions between different types of actors is also necessary.

Content feature mining (addressing RQ2)

Emotion analysis

The distribution of emotion scores for various actor types was analyzed (Figure 2). Inauthentic actors, such as astroturfing, exceptional users, news, and bots, showed higher negative emotion scores than positive emotions. Astroturfing had the highest negative emotion score (median = 0.040), followed by exceptional users (median = 0.037), news (median = 0.035), and bots (median = 0.032). In contrast, the highest positive emotion scores were in politics (median = 0.028) and NGOs & activists (median = 0.0279). Positive emotion scores for inauthentic actors were lower: astroturfing (median = 0.024), exceptional users (median = 0.023), and bots (median = 0.020). Further analysis of negative emotions (Figure 3) revealed that astroturfing and news had the highest scores in categories like disgust, fear, anger, and sadness. For example, exceptional users had a median disgust score of 0.01, while astroturfing scored 0.009.

Figure 2.

Distribution of emotion scores.

Figure 3.

Distribution of negative emotion median scores.

Therefore, for RQ2, it can be concluded that negative sentiment was more prominent in inauthentic actors and news, while positive sentiment showed a different pattern.

Actor-network analysis (addressing RQ3)

Social network analysis

The emotion score distribution showed that inauthentic actors were more likely to spread negative emotions. To observe emotional contagion, the study used social network analysis of retweet relationships for RQ3.

Figure 4 illustrates that astroturfing accounts amplified negative emotions from news and political actors, which were then transmitted to normal users. News, astroturfing, and political actors had the highest weighted out-degree, mainly acting as senders of negative emotions (Table 5). Astroturfing accounts not only sent negative messages (weighted out-degree = 109,504) but also received the most negative messages (weighted in-degree = 344,726), followed by normal users (weighted in-degree = 121,314). Astroturfing accounts became a major transit node for negative information. Overall, news and political actors, astroturfing accounts, and normal users were key nodes, with their dynamic interactions warranting further observation.

Figure 4.

Retweet the relationship between negative words. The weighted in-degree value of a node was represented as the node size, and the frequency of negative words in retweeted tweets was represented as the thickness and transparency of the edge.

Table 5.

Weighted In-degree and out-degree of actor node.

Actors	Weighted in-degree	Weighted out-degree
News	2136	204,354
Astroturfing	344,726	109,504
Politics	310	79,681
Normal users	121,314	59,245
NGOs & activists	192	15,496
Exceptional users	52,843	26,624
Bots	20,665	9663
Sports	52	1718
Business	231	7441
Academics	122	2947
Entertainment	275	26,193

Emotional contagion analysis (addressing RQ4)

Time series analysis

To analyze the emotional contagion between key actors, Granger causality tests were conducted on the time series of daily tweets, with results shown in Tables 6 and 7.

Table 6.

VAR Granger causality test (negative emotion).

Cause	Caused	Wald test statistic χ2	Critical value	p value
Astroturfing	Normal users	40.45	18.31	.000
Exceptional users	Normal users	19.22	18.31	.038
Bots	Normal users	14.99	18.31	.132
News	Normal users	23.17	18.31	.010
Politics	Normal users	18.39	18.31	.049
NGOs & activists	Normal users	15.07	18.31	.130
Normal users	Astroturfing	45.00	18.31	.000
Normal users	Exceptional users	37.85	18.31	.000
Normal users	Bots	15.54	18.31	.114
Normal users	News	22.94	18.31	.011
Normal users	Politics	18.16	18.31	.052
Normal users	NGOs & activists	24.68	18.31	.006

Table 7.

VAR Granger causality test (positive emotion).

Cause	Caused	Wald test statistic χ2	Critical value	p value
Astroturfing	Normal users	8.423	18.31	.588
Exceptional users	Normal users	17.26	18.31	.069
Bots	Normal users	11.48	18.31	.321
News	Normal users	20.52	18.31	.025
Politics	Normal users	5.123	18.31	.883
NGOs & activists	Normal users	32.46	18.31	.000
Normal users	Astroturfing	10.72	18.31	.379
Normal users	Exceptional users	19.41	18.31	.035
Normal users	Bots	14.15	18.31	.166
Normal users	News	25.85	18.31	.004
Normal users	Politics	13.47	18.31	.198
Normal users	NGOs & activists	14.45	18.31	.154

The findings indicated that the negative emotions of inauthentic and political actors significantly Granger caused the negative emotions of normal users (Table 6). For example, significant relationships were found between Astroturfing and Normal users (χ2 = 40.45, p = .00), Exceptional users and Normal users (χ2 = 19.22, p = .038), and Politics and Normal users (χ2 = 18.39, p = .049). In contrast, no significant Granger causality was found for positive emotions between these actors and normal users (Table 7), such as between Astroturfing and Normal users (χ2 = 8.42, p = .588) or Politics and Normal users (χ2 = 5.12, p = .883). This indicates that negative emotions from inauthentic and political actors significantly influenced normal users, while positive emotions did not.

Additionally, normal users also had reverse emotional contagion effects on astroturfing accounts (χ2 = 45, p = .000), exceptional users (χ2 = 37.85, p = .000), news (χ2 = 22.94, p = .011), and NGOs & activists (χ2 = 24.68, p = .006) (Table 6). Inauthentic actors could capture and amplify the negative emotions of normal users, creating a closed loop of emotional contagion.

Discussion

In summary, the key findings of this study are: (1) Multiple actors spread misinformation, with hybrid human-machine astroturfing accounts and bots playing a significant role, mainly spreading misinformation through retweets and likes, while influential users in news were more active in posting original tweets. (2) Inauthentic actors (astroturfing accounts, bots, exceptional users) and news actors were more likely to spread negative misinformation. (3) In the retweeting network, astroturfing accounts amplified the negative emotions of news and political users, which were then transmitted to normal users. (4) Negative emotions from inauthentic actors and political users were Granger causes of negative emotions in normal users, while positive emotions were not significant.

This study examines the mechanism of digital emotion contagion in health misinformation, particularly driven by hybrid human-machine astroturfing accounts. While prior research has extensively documented the role of bots in disseminating COVID-19 misinformation,^5,7,27 the complex interplay between human and machine actors in computational propaganda is often overlooked. As a study demonstrated, false news spreads farther and faster than the truth, partly due to its emotional novelty.²⁸ Building on these findings, our analysis suggests that hybrid coordination may exacerbate this effect by strategically amplifying negative emotions, thereby accelerating misinformation diffusion. Additionally, this study highlights the importance of integrating various computational methods for misinformation research. Tools like Botometer have limitations in identifying new users in specific events,²⁹ whereas the astroturfing collaborative networks method offers a more comprehensive view of digital emotion contagion. This underscores the need for future studies to select flexible and appropriate computational methods based on their research objectives.

Digital emotion contagion is mediated by platform biases and organized misinformation, often amplifying strong negative emotions. This misinformation can lead to public psychological reactions such as panic, incitement, and exclusionary behaviors (e.g., discrimination), harming public communication. It also contributes to stigma, hinders health measures adoption, and fuels distrust in authorities, potentially causing mass panic.³⁰ Additionally, the politicization of health issues, especially under right-wing leaders like Donald Trump, has been heavily influenced by computational propaganda and affective politics.³¹ The study’s findings offer a framework for understanding the spread of misinformation in multi-agent networks and its implications for health issues.

While this study primarily focuses on misinformation and emotional contagion on Twitter, it is important to acknowledge that these phenomena extend to other digital platforms, particularly those with younger and more diverse user bases. For instance, TikTok has emerged as a significant vector for emotionally charged content, including medical misinformation, especially in non-English-speaking regions.³² Future research should explore these dynamics in other digital ecosystems to develop more comprehensive strategies for mitigating misinformation.

Limitation

This study has several limitations. First, while it focused on online digital emotion contagion, particularly in digital health monitoring, the short- and long-term psychological effects on users require further examination. Second, as the analysis relied solely on Twitter data, platform-specific algorithms may have shaped contagion patterns; cross-platform comparisons would strengthen generalizability. Third, the emotional analysis was also restricted to Twitter text, leaving multimodal sentiment analysis for future exploration, including video and image data. Finally, our analysis did not disentangle platform-level emotional amplification from inter-actor contagion dynamics, though the latter remains salient even under uniform visibility conditions. Future work should integrate platform affordances with network-level actor strategies.

Conclusion

This study develops a computational framework to explore digital emotional contagion in health misinformation, focusing on hybrid human-machine astroturfing accounts. Astroturfing accounts acted as key transit nodes, creating a closed loop of negative emotional contagion in misinformation spread. The framework provides a tool for monitoring emotional responses in public health crises, with findings applicable to other emergencies beyond COVID-19.

Footnotes

ORCID iDs

Qiuyi Chen

Qian Liu

Ethical considerations

This study involved the analysis of secondary data collected from social media platforms. The data set includes publicly available information from social media accounts. As such, ethics approval was not sought. The results are presented in aggregate form and do not contain any identifiable information, ensuring the privacy of the individuals whose data were included in the study.

Author contributions

CQY conceptualized the study, collected and analyzed the data, designed visualizations, and drafted the manuscript with subsequent revisions. LQ supervised the research project, provided strategic direction, reviewed and provided critical feedback on manuscript drafts, and approved the final version of the paper. Both authors contributed to refining the content and ensuring the accuracy of the research.

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research was supported by Key-Area Research and Development Program of Guangdong Province; ID:2022B101010004; Project of Collaborative Innovation Centre for Chinese Culture Transmission and Communication in Hong Kong, Macao, Taiwan and Overseas, Jinan University; JNXT2022002; Chinese Special Funds for Basic Research Operating Costs of Central Universities; ID:23NJYH11.

Declaration of conflicting interests

The authors declare that there are no conflicts of interest regarding the publication of this paper.

Data Availability Statement

The data that support the findings of this study are available from the corresponding author upon reasonable request. Due to privacy and ethical considerations, some data may not be publicly available.*

References

Pummerer

Winter

Sassenberg

. Addressing Covid-19 vaccination conspiracy theories and vaccination intentions. J Health Commun 2022; 3(2): 1–12.

Wang

Walter

Ophir

. Not all bots are created equal: the impact of bots classification techniques on identification of discursive behaviors around the COVID-19 vaccine and climate change. Soc Sci Comput Rev 2024; 42(2): 394–415.

Velásquez

Leahy

Restrepo

, et al. Online hate network spreads malicious COVID-19 content outside the control of individual social media platforms. Sci Rep 2021; 11(1): 11549.

Ugarte

. The rise of anti-Asian hate in the wake of Covid-19 [Internet]. Items 2020. [cited 2021 Sep 15]. Available from: https://items.ssrc.org/covid-19-and-the-social-sciences/the-rise-of-anti-asian-hate-in-the-wake-of-covid-19/

Himelein-Wachowiak

Giorgi

Devoto

, et al. Bots and misinformation spread on social media: implications for COVID-19. J Med Internet Res 2021; 23(5): e26933.

Schoch

Keller

Stier

, et al. Coordination patterns reveal online political astroturfing across the world. Sci Rep 2022; 12(1): 4572.

Jemielniak

Krempovych

. An analysis of AstraZeneca COVID-19 vaccine misinformation and fear mongering on Twitter. Public Health 2021; 200: 4–6.

Herrando

Constantinides

. Emotional contagion: a brief overview and future directions. Frontiers in Psychology [Internet] 2021; 16: 712606[cited 2022 Sep 12];12. Available from: https://www.frontiersin.org

Kramer

ADI

Guillory

Hancock

. Experimental evidence of massive-scale emotional contagion through social networks. Proc Natl Acad Sci USA 2014; 111(24): 8788–8790.

10.

Goldenberg

Gross

. Digital emotion contagion. Trends Cognit Sci 2020; 24(4): 316–328.

11.

Stella

Ferrara

Domenico

. Bots increase exposure to negative and inflammatory content in online social systems. Proc Natl Acad Sci USA 2018; 115(49): 12435–12440.

12.

Hagen

Neely

Keller

, et al. Rise of the machines? Examining the influence of social bots on a political discussion network. Soc Sci Comput Rev 2020, 089443932090819.

13.

Kušen

Strembeck

. Something draws near, I can feel it: an analysis of human and bot emotion-exchange motifs on Twitter. Online Social Networks and Media 2019; 10–11: 1–17.

14.

Cai

Luo

Meng

, et al. Network distribution and sentiment interaction: information diffusion mechanisms between social bots and human users on social media. Inf Process Manag 2023; 60(2): 103197.

15.

Muñoz

Díez

Bellogín

. Modeling disinformation networks on Twitter: structure, behavior, and impact. Appl Netw Sci 2024; 9(1): 1–35.

16.

Unlu

Truong

Sawhney

, et al. Setting the misinformation agenda: modeling COVID-19 narratives in Twitter communities. New Media Soc 2024; 27(7): 3973–3997.

17.

Zhang

Chen

Lukito

. Network amplification of politicized information and misinformation about COVID-19 by conservative media and partisan influencers on Twitter. Polit Commun 2023; 40(1): 24–47.

18.

The CoronaVirusFacts/DatosCoronaVirus alliance database

[Internet]. Poynter. [cited 2020 Jun 10]. Available from: https://www.poynter.org/ifcn-covid-19-misinformation/

19.

Features

. spaCy usage documentation. [Internet]. Linguistic Features, [cited 2020 Feb 15]. Available from: https://spacy.io/usage/linguistic-features#pos-tagging

20.

Řehůřek

Sojka

. Software framework for topic modelling with large Corpora In: Proceedings of the LREC 2010 Workshop on New Challenges for NLP Frameworks. Malta: ELRA. 2010; 45–50.

21.

Mikolov

. Distributed representations of sentences and documents [Internet]. arXiv:1405.4053 [cs] 2014, [cited 2020 Sep 2]. Available from: https://arxiv.org/abs/1405.4053

22.

Vrandečić

Krötzsch

. Wikidata: a free collaborative knowledgebase. Commun ACM 2014; 57(10): 78–85.

23.

Keller

Schoch

Stier

, et al. Political astroturfing on Twitter: how to coordinate a disinformation campaign. Polit Commun 2020; 37(2): 256–280.

24.

Yang

Varol

Davis

, et al. Arming the public with artificial intelligence to counter social bots. Hum Behav Emerg Technol 2019; 1(1): 48–61.

25.

Mohammad

Turney

. Emotions evoked by common words and phrases: using mechanical turk to create an emotion lexicon. Proceedings of the NAACL HLT 2010 Workshop on Computational Approaches to Analysis and Generation of Emotion in Text [Internet]: Association for Computational Linguistics, 2010, pp. 26–34. [cited 2020 Apr 20]. Available from: https://www.aclweb.org/anthology/W10-0204

26.

Chang

Jayne

. A systematic review of social media-based sentiment analysis: emerging trends and challenges. Decision Analytics Journal 2022; 3: 100073.

27.

LHX

Kloo

Clark

, et al. An exploratory analysis of COVID bot vs human disinformation dissemination stemming from the disinformation dozen on telegram. J Comput Soc Sci 2024; 7: 695–720, [cited 2024 Jun 8]; Available from: https://doi.org/10.1007/s42001-024-00253-y

28.

Vosoughi

Roy

Aral

. The spread of true and false news online. Science 2018; 359(6380): 1146–1151.

29.

Beatson

Gibson

Cunill

, et al. Automation on Twitter: measuring the effectiveness of approaches to bot detection. Soc Sci Comput Rev 2023; 41(1): 181–200.

30.

Shigemura

Harada

Tanichi

, et al. Rumor-related and exclusive behavior coverage in internet news reports following the 2009 H1N1 influenza outbreak in Japan. Disaster Med Public Health Prep 2015; 9(4): 459–463.

31.

Boler

Davis

. The affective politics of the “post-truth” era: feeling rules and networked subjectivity. Emotion, Space and Society 2018; 27: 75–85.

32.

Izquierdo-Condoy

Arias-Intriago

Mosquera-Quiñónez

, et al. Assessing the educational impact and quality of medical microvideos on TikTok: the case of Latin America. Med Educ Online 2025; 30(1): 2474129.

Unveiling emotional contagion in COVID-19 misinformation: Computational analysis for public health crisis surveillance

Abstract

Keywords

Introduction

Methods

Data collection and preprocessing

Actor feature mining (addressing RQ1)

Knowledge graph and entity analysis

Social network analysis and bot detection

Content feature mining (addressing RQ2)

Emotion analysis

Actor-network analysis (addressing RQ3)

Social network analysis

Emotional contagion analysis (addressing RQ4)

Time series analysis

Results

Actor feature mining (addressing RQ1)

Knowledge graph and entity analysis

Social network analysis and bot detection

Content feature mining (addressing RQ2)

Emotion analysis

Actor-network analysis (addressing RQ3)

Social network analysis

Emotional contagion analysis (addressing RQ4)

Time series analysis

Discussion

Limitation

Conclusion

Footnotes

ORCID iDs

Ethical considerations

Author contributions

Funding

Declaration of conflicting interests

Data Availability Statement

References