Abstract
China is home to the world’s largest population, with the potential for disease outbreaks to affect billions. However, knowledge of Chinese vaccine acceptance trends is limited. In this work we use Chinese social media to track responses to the recent Changchun Changsheng Biotechnology vaccine scandal, which led to extensive discussion regarding vaccine safety and regulation in China. We analyzed messages from the popular Chinese microblogging platform Sina Weibo in July 2018 (n=11,085), and August 2019 (n=500). Thus, we consider Chinese vaccine acceptance, before, during, immediately after, and one year after the scandal occurred. Results show that expressions of distrust in government pertaining to vaccines increased significantly during and immediately after the scandal. Self-reports of vaccination occurred both before, and one year after, the scandal; however, these self-reports changed from positive endorsements of vaccination to concerns about vaccine harms. Data suggest that expressed support for vaccine acceptance in China may be decreasing.
Keywords: Vaccine Acceptance, Vaccine Safety, Weibo, Social Media, Changchun Changsheng, China Vaccine Incident
Introduction
Vaccine acceptance is a crucial public health issue (Dubé, Vivion, & MacDonald, 2015), which has been exacerbated by the use of social media to spread content opposing vaccination (Betsch et al., 2012). However, data from social media have also provided new insights into the drivers of vaccine acceptance. Beyond supplementing traditional measures of vaccine hesitancy (Dredze, Broniatowski, Smith, & Hilyard, 2015), studies using data from Twitter (e.g., Broniatowski et al., 2018, who found evidence that state-sponsored and pecuniary interests amplify online vaccine debates to promote discord and distribute spam and malware), Facebook (e.g., Orr, Baram-Tsabari, & Landsman, 2016, who found that polio vaccine refusal was associated with religious rationales), and Reddit (e.g., Lama, Hu, Jamison, Quinn, & Broniatowski, 2019, who found that misconceptions regarding HPV prophylaxis, but not sexual promiscuity, were spread in online communities) have shown that social media can provide new information regarding the dynamics of vaccine communication online, potentially affecting real-world vaccine uptake behaviors. China has the world’s largest population, with several cities among the world’s top population densities. Thus, disease outbreaks in China have the potential to affect billions of people. However, little is known of vaccine acceptance in China on the national level. The vast majority of the Chinese population is vaccinated: Cao et al (2012) conducted a national level survey in China (4,681 children from 32 counties) of the coverage of the Expanded Program on Immunization (EPI) vaccines. This study found that primary immunization coverage of original EPI vaccines was overall above 90% and the coverage of new vaccines integrated into EPI after 2008 was overall above 85. However, such studies do not directly measure public attitudes towards vaccination.
Social media may be used to measure public opinion regarding vaccination in a manner that is both fast and relatively uncorrupted by response bias due to the self-reported nature of social media messages (Dredze, Broniatowski, Smith, & Hilyard, 2016). However, few studies have been carried out examining Chinese social media due to practical and linguistic barriers: Most people residing in mainland China cannot access international social media without a Virtual Private Network (VPN) service. As a result, most Chinese citizens rely on China-based social media. Sina Weibo is well-known as the Chinese version of Twitter, although it has more users than Twitter does (Radcliffe, 2017). In this study we examine Weibo data to better understand trends in Chinese vaccine hesitance. Specifically, we examine the fallout of the 2018 Changchun Changsheng Biotechnology vaccine scandal and ensuing confidence crisis (The Lancet, 2018).
2018 Vaccine Incident in China
On July 15th, 2018, several Chinese newspapers reported that government inspectors had determined that Changchun Changsheng Biotechnology, a prominent manufacturer of vaccines in China, had violated national regulations and standards when producing 250,000 rabies vaccine doses (Tan, 2018). Later that day, this incident was confirmed by The China Food and Drug Administration. The violation might have significantly undermined the effectiveness of the involved rabies vaccines. However, no one reported any adverse reactions or deaths from the involved vaccines before they were recalled in July, 2018 (Hong & Wei, 2018). News of this incident slowly escalated on Chinese social media platforms (Figure 1).
Figure 1.
Frequency of collected Messages with related keywords
On July 21st, 2018, one blog article critiquing Changsheng Biotechnology, “The King of Vaccines,” became the most popular article of the day. On July 23rd, 2018, the article peaked with 2 million shares, in which the author directly referred to the vaccine incident as a major scandal. Meanwhile, Chinese people voiced their concerns on traditional media, advertisements, and in patriotic protests during the incident. The Chinese government acted quickly after the incident (Hong & Wei, 2018) leading Changsheng Biotechnology to declare bankruptcy on June 27th, 2019.
Research Question
The Changsheng Biotechnology incident seems to have had lasting impacts on vaccine acceptance in China. For example, Liu et al. conducted a cross-sectional study using another Chinese social media platform – WeChat – to a sample that was “relatively well-educated, had higher incomes and lived in urban areas”, finding that a majority of their sample held negative attitudes toward vaccines in September 2018. We aim to extend this cross-sectional work with a longitudinal analysis of Weibo messages pertaining to vaccines. Specifically, this paper aims to compare the vaccine discourse on Weibo immediately before the incident to those immediately after and one year after the incident, thus providing insight into online expressions of vaccine acceptance in the short- and long-terms.
Methods
Data Collection
Discussion regarding the 2018 vaccine incident largely unfolded on Weibo. Thus, we conducted a retrospective observational study using a convenience sample from this social medium as our primary data source. We crawled public messages on Weibo.com with the keyword 疫苗 (Chinese for “vaccine”) from July 15th, 2018 to July 31st, 2018 yielding 22 million messages. We downloaded 11,085 of these messages using Weibo’s API.1 We performed a similar search in August, 2019, yielding 10.5 million messages. Due to changes in Weibo’s API, we manually collected another 500 messages in 2019 to assess the long-term effect of the scandal.
Messages were categorized into three time periods:
Pre-Incident Period, 7/1/2018–7/10/2018. During this period, the incident was still under investigation, and had not yet been reported to the public. We assume that data in this period is representative the vaccine-related discussion on Weibo immediately prior to the incident. 2,721 messages were collected in total for this period.
Escalating Period, 7/11/2018–7/20/2018. During this time period, an investigation was carried out and reported by several minor national newspapers. At this time, more and more people became aware of the severity of the vaccine incident. 3,808 messages were collected in total for this period.
National Headline Period, 7/21/2018–7/31/2018. In this period, the incident became one of the most controversial topics in China. 4,557 messages were collected in total for this period.
The following table has summarized the time-period distribution of the collected messages.
Study 1: Qualitative Analysis of Messages During the 2018 Vaccine Scandal
To answer our first research question, we assigned 500 randomly selected Weibo messages into five categories using qualitative annotation: (1) Animals/Pet, messages referring to pet and animal vaccines; (2) Scandal/Incident, messages about the vaccine scandal; (3) Harm/Injury, reports on adverse vaccine reactions; (4) Trust, messages that debate societal or moral issues, or expresses frustration that faith in the government might have been misplaced; (5) Self-Reporting Vaccination Experience, explicit self-reports about vaccination or intention to vaccinate. The key difference between categories 3 and 5 is that the messages from category 3 report some forms of adverse reaction after receiving a vaccine and messages from category 5 do not. Messages that did not correspond to any of these categories were marked Unsure, with specific notes. This qualitative codebook was prepared and reviewed by all four authors (See Supplement 2 for coding rules and examples messages of each category). All annotations can be found in Supplement 3.
Two authors (DH & CM), both of whom read Chinese fluently, categorized all 500 messages independently using the codebook. When messages were categorized as Unsure one of the paper’s authors (CM) made notes describing this annotation.
We also carried out a second analysis of another 500 messages in 2018 to verify several unexpected observations found in the first round of annotation.
Study 2: Qualitative Analysis of Messages One Year After the Scandal
To assess the long-term effects of the scandal, we carried out the same search from August 1st, 2019 to August 10th, 2019, yielding another 500 messages. Before we began this collection, Weibo’s user agreement changed to require log-in credentials for the aforementioned web-based API. The new policy made it impossible for us to access the same amount of data in 2019 as in 2018. Therefore, we only collected 500 messages by manually downloading 25 webpages generated by the API. The same coders (DH & CM) categorized all 500 messages independently using the same codebook for Study 1.
Study 3: Topic Model Analysis
In order to verify that our qualitative coding results were representative of the full set of 11,585 messages extracted, we used Latent Dirichlet Allocation (LDA) (Blei, Ng, & Jordan, 2003), a probabilistic topic model, to automatically segment all Weibo messages into 20 probabilistic “topics”. Specifically, we used unigram and bigram features with hyperparameter values of alpha=0.1 and η=0.01. When applied to a corpus of messages, LDA returns the probability that each message is about each of these topics. Although LDA output requires interpretation by humans, the result can be used to provide an overall description of the whole corpus of messages.
Results
After the scandal incident, self-reported vaccination decreases and discussion of trust in government increases
For Study 1, the two annotators independently reached the same labels in all but three cases out of 500 with high reliability (Cohen’s Kappa = 0.98). These three differences were rooted in ambiguous names of pets and were soon resolved by discussion. The result of the first analysis is shown in Figure 1.
There were significant differences between the first- and second periods of our analysis, χ2(5)=41.60, p<0.001, and between the second and third periods, χ2(5)=224.86, p<0.001. Figure 1 shows that, before the incident, 24% of messages described self-reported vaccination. However, this category dropped to 14%, during the Escalating Period, and it eventually became 1%, in the third period. In contrast, before the incident, very few people on Weibo had discussed trust in government in the context of vaccines, but this number reached a high of 54% by the third period.
Messages not relevant to the incident, but expressing concerns of vaccine harm or injury were only observed five times during the Pre-Incident Period; however, after the incident, twelve messages that were mainly about trust also explicitly used the word “harm people”, “harm baby” or “harmful”. Finally, messages discussing animal/pet related vaccines accounted for 30% of all messages during the Pre-Incident Period but eventually vanished to 0%, by the third period.
Among the 122 messages categorized as “unsure”, we found that 85 messages contained pro-vaccine content, including advertisements for vaccines by local clinics, advertisements for non-mandatory vaccines, or educational material for the public about the vaccine schedule for newborns. Out of these 85 pro-vaccine messages, 14 of them promoted vaccination trips to Hong Kong, or identified themselves as vaccine providers located in Hong-Kong, where the healthcare standards are considered higher by most Chinese people.
Secondary analysis: Rates of pro-vaccine advertisement and self-reported vaccination decreased
The slightly unexpected finding from the first analysis led us to carry out a second analysis to verify several of our observations. Specifically, we aimed (1) to assess whether pro-vaccine advertisement rates had declined after the incident, and (2) to assess whether self-reporting of vaccination had declined. Thus, we pulled another 500 samples from the original 11,085 messages and annotated with the same coding procedure.
In this second sample, we found that 61% of messages in the first period contained pro-vaccine advertisements; however, this percentage dropped to 12% by the third period. We also witnessed a decline in the number of messages self-reporting vaccination, χ2(1)=13.33, p<0.001.
All self-reporting messages before the scandal were neutral or positive in sentiment. In total, only seven messages directly expressed a negative sentiment when self-reporting vaccination, and all seven were reported after the vaccine incident was exposed by Chinese government agencies.
Study 2: One Year Later, Self-Reported Vaccine Harms Have Doubled
In this analysis, the two annotators’ agreement was only moderate (Cohen’s Kappa = 0.53). The disagreement resulted from two types of messages: one author interpreted ambiguous messages using the word “baby” (in Chinese : 宝宝) as referring to humans, whereas the other interpreted these messages as referring to pets. After discussion, we choose to interpret the word “baby” as referring to human babies. A second source of disagreement came from discrepancies as to whether some self-reports should be categorized as harm/injury because of their negative sentiment about vaccines. We did not categorize a message as expressing vaccine harms unless it did so explicitly.
These results show that the proportion of messages reporting adverse reactions to vaccines or expressing concerns of potential vaccine harm jumped from 4.5% to 14.42% in the past year. Compared to the pre-incident period in Study 1, there was a significant change, χ2(1) = 5.795, p=0.016 in how messages pertaining vaccination were distributed, with a marked increase in the proportion of messages self-reporting vaccination. Furthermore, the tone of these messages changed from primarily positive to primarily negative. However, the scandal itself was rarely discussed.
Results Replicate Across Entire Corpus
We fit a 20-topic model to the 11,515 messages in our corpus and manually aggregated the output into 6 overarching topics based on the most representative words in each topic. Table 2 displays the top 5 tokens (either unigram or bigram) of each topic in Chinese, their English translations and to which of the 6 aggregate topics it was assigned to. The data and other details of the LDA model can be found in Supplement 4.
Table 2.
Trends of pro-vaccine ads and self-reporting, July 2018
| Pro Vaccine Ads (%) | Self-Reporting (%) | |
|---|---|---|
| July 1 - July 10 (n=168) | 103 (61) | 30 (18) |
| July 11- July 20 (n=155) | 84 (54) | 31 (20) |
| July 21- July 31 (n=177) | 22 (12) | 2 (1) |
We determined the final 6 topics to be: (1) Vaccine Ads, (2) General News/Education/Discussion Related To Vaccine, (3) Scandal Development, (4) Distrust and Complaints against Government, (5) Self-Report Experience, and (6) Pet vaccination. Figure 3 demonstrates how the distribution of these 6 topics developed overtime. Four of our main findings from prior sections are in qualitative agreement with the LDA model’s output: First, we can see the rapid development of the scandal in late July, 2018. Second, pro-vaccine advertisements gradually decreased as the incident developed, and in 2019 it remained lower than the pre-incident level of 2018. Third, in 2019, people on Weibo were more likely to self-report vaccination (and potentially, vaccine harms) compared to the pre-incident level of 2018. Finally, we found that in 2019, people were more likely mention the Chinese government when mentioning vaccines: a rarity before the incident.
Figure 3.
Shift of Distribution of 6 categories, Pre-incident Data vs. 2019 Data
Discussion
Our results demonstrate a shift in the vaccine discourse on Weibo. Immediately after the incident, the frequency of self-reported vaccination declined significantly, while the discussion of trust in government increased. One year later, discussions of trust in government are no longer present and self-report rates increased significantly. However, the nature of these self-reports changed. Before the incident, self-reports were primarily positive in nature, whereas one year after the incident, vaccinators expressed more concerns. Furthermore, a higher proportion of messages reported harms from vaccination.
Our results indicate that the experience of vaccination has become a more conspicuous topic among Weibo users, with many actively seeking information on this issue. This may suggest the erosion of vaccine confidence in Chinese society, since prior research has shown that the very presence of a vaccine debate can lead to more vaccine hesitancy (Dixon & Clarke, 2013).
Nine types of vaccines are virtually mandatory in China due to the “no vaccine, no education” policy2. Nevertheless, an erosion of vaccine confidence in other countries has led to a worldwide environment where vaccine policy is routinely questioned, with some countries failing to enforce mandates, introducing weak mandates (Attwell et al., 2018), or even overturning existing mandates (Gollust, Dempsey, Lantz, Ubel, & Fowler, 2010; Kennedy, 2019). Thus, the potential emergence of vaccine opposition in China is a potential cause for concern, especially considering the density of several large Chinese population centers. Our analysis thus presents a cautionary tale. A single, significant vaccine incident in China appears to have had a lasting effect on the population, potentially increasing the public health risk. This study highlights the dangers of public perception of even a single vaccine safety incident.
Limitations and Directions for Future Work
Our research questions are not intended to draw any conclusions regarding persons or organizations who might be responsible for the incident, nor can we evaluate the performance of the Chinese government when handling the incident or banning Weibo messages, or the extensive efforts of Chinese medical professionals, who often spend decades fighting a number of infectious diseases in dangerous and underpaid conditions. Thus, although we study a major social incident in China, we explicitly do not take a political stance vis-à-vis the actions of the Chinese government, nor can we draw conclusions regarding the efficacy of these actions.
Our study is retrospective and observational, and therefore does not allow us to draw causal conclusions regarding the effect of the incident on vaccine acceptance in China. Thus, one might argue that other events pertaining to vaccines might have caused the observed changes. For example, incidents in Shanxi in 2006 and 2008 (China.org.cn, 2010) and Shandong in 2016 (BBC, 2016) might have also influenced the vaccine discussions in China. However, we did not encounter any messages that mentioned these prior events. Furthermore, our study used an extended time-series design to compare the time period immediately before the scandal to the time periods immediately after, and one year after, it. Our data indicate clearly that the events of July 2018 were associated with significant changes in the online discourse. Thus, it is unlikely that other events might have caused the trends observed.
Our analysis relied on data from Sina Weibo – one of the largest Chinese social media platforms. In general, studying Weibo data presents several challenges: Compared to Twitter, Weibo’s data are harder for researchers to access – per Weibo’s policy, businesses or organization in China can pay for access to an “advanced API,” which enables “comprehensive search [of] recent messages based on the keywords”. Thus, according to Weibo’s policies, a small number of Chinese organizations can obtain special access to a comprehensive sample of Weibo data. These permissions are determined on a case by case basis with a strong bias against international applicants. Furthermore, the Chinese government sometimes makes determinations regarding which content may be posted on Weibo. Additionally, Weibo has to comply with Chinese regulations: several government agencies in China can issue legal orders requiring Weibo to remove all messages with certain keywords or hashtags. Weibo can also hide certain messages from appearing in search queries. Ultimately, we do not know the details of which messages might be censored using this legal mechanism. However, research (Fu, Chan, & Chau, 2013; Xu, Mao, & Halderman, 2011) has shown that certain keywords are censored on Weibo, with more aggressive censorship at certain sensitive times. The advanced API is not accessible to researchers residing in the U.S. Thus, we were limited to the messages available from the Weibo web response, and we do not have information from Sina.com regarding how the web response is sampled or ranked. Our results nevertheless demonstrate clear differences between the time periods in our analysis. In particular, the presence of a significant number of messages expressing concerns of vaccine harm one year after the scandal (but not one year before the scandal) seems to suggest that these messages are not being censored in an obvious way and so may represent actual changes in Chinese sentiment towards vaccination. If so, then the observations in this paper can be an indication that public perception of vaccination may now include risks of harm from the vaccine and an overall decrease in vaccine confidence.
Supplementary Material
Figure 2.
Trends of 6 categories in July 2018
Figure 4.
6 Aggregated Topics from LDA Model
Table 1:
Time-period distribution of 2018 data
| Total | Study 1, First Analysis | Study 1, Secondary Analysis | |
|---|---|---|---|
| July 1 - July 10 | 2721 | 111 (4%) | 168 (6%) |
| July 11- July 20 | 3808 | 174 (5%) | 155 (4%) |
| July 21- July 31 | 4557 | 216 (5%) | 177 (4%) |
Table 3.
LDA model summary: Assigned Category, Top Keyword in Chinese, and their English Translation
| ID | Category | Top 5 Chinese Words | English Translation | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| 0 | Vaccine Ads | hpv | 疫苗 | 香港 | hpv疫苗 | 宫颈癌 | Hpv | Vaccine | Hong kong | Hpv-vaccine | Cervical Cancer |
| 1 | Scandal Development | 生产 | 疫苗 | 违法 | 长生 | 狂犬病 | Production | Vaccine | Illegal | Changsheng | Rabies |
| 2 | General News/Education/Dis cussion Related To Vaccine | 疫苗 | 接种 | 展开 | 展开全文 | 全文 | Vaccine | Inoculation | Expand | Expand-Full Article | Full Article |
| 3 | Scandal Development | 疫苗 | 武汉 | 百白破 | 生物 | 武汉生物 | Vaccine | Wuhan | DPT | Biology | Wuhan-Bioloy |
| 4 | Self-Report Experience | 疫苗 | 孩子 | 注射 | 注射疫苗 | 感冒 | Vaccine | Children | Injection | Inject-Vaccine | Flu |
| 5 | Scandal Development | 事件 | 疫苗 | 疫苗事件 | 关注 | 粉丝 | Incident | Vaccine | Vaccine-incident | Follow | Fan |
| 6 | Scandal Development | cn | http | http-cn | 疫苗 | 新闻 | cn | http | http-cn | Vaccine | News |
| 7 | Pet vaccination | 领养 | 疫苗 | xe627 | 驱虫 | 展开 | Adoption | Vaccine | Unicode xe627 (V) | Worm treatment | Expand |
| 8 | General News/Education/Dis cussion Related To Vaccine | 伤口 | 狂犬 | 狂犬疫苗 | 冲洗 | 疫苗 | Wound | Rabies | Rabies-Vaccine | Wash | Vaccine |
| 9 | Scandal Development | 生产 | 疫苗 | 长生 | 狂犬疬 | 疫苗生产 | Production | Vaccine | Changsheng | Rabies | Vaccine-Production |
| 10 | Scandal Development | 生物 | 长生 | 长生生物 | 疫苗 | 造假 | Biology | Changsheng | Changsheng-Biology | Vaccine | Fraud |
| 11 | Self-Report Experience | 疫苗 | 今天 | 没有 | 自己 | 时候 | Vaccine | Today | No | Self | Time |
| 12 | Scandal Development | 长春 | 长生 | 长春长生 | 疫苗 | 狂犬病 | Changchun | Changshen | Changchun-Changshen | Vaccine | Rabies |
| 13 | Distrust and Complain against Government | 疫苗 | 我们 | 没有 | 什么 | 一个 | Vaccine | We | Don’t have | What | One |
| 14 | Scandal Development | http | cn | http-cn | 疫苗 | 狂犬 | http | cn | http-cn | Vaccine | Rabies |
| 15 | Distrust and Complain against Government | 疫苗 | 问题 | 中国 | 监管 | 企业 | Vaccine | Problem | China | Regulation | Cooperation |
| 16 | Vaccine Ads | 围观 | 免费 | 免费围观 | 一起 | 问题 | Broadcasting | Free | Free-Broadcasting | Together | Problem |
| 17 | Self-Report Experience | 疫苗 | 出现 | 宝宝 | 可以 | 大家 | Vaccine | Appear | Baby | Could | Everyone |
| 18 | Self-Report Experience | 宝宝 | 疫苗 | 接种 | 检查 | 观察 | Baby | Vaccine | Inoculation | Check | Observe |
| 19 | General News/Education/Dis cussion Related To Vaccine | 疫苗 | 接种 | 接种疫苗 | 有效 | 注射 | Vaccine | Inoculation | Inoculation-Vaccine | Effective | Injection |
Highlights.
We use Weibo to study vaccine acceptance in China after the 2018 vaccine scandal
Messages pertaining to vaccines expressed more distrust of government after the scandal
Sentiment of self-reports of vaccination changed from positive to concerned
Long-term effect of the scandal shows vaccine acceptance in China may be decreasing
Acknowledgments
This research was supported by the National Institute of General Medical Sciences, National Institutes of Health (NIH; award 5R01GM114771). Note: The content is solely the responsibility of the authors and does not necessarily represent the official views of NIH
Footnotes
Conflict of interest statement
Dr. Dredze reported receiving personal fees from Bloomberg LP and Good Analytics outside the submitted work. No other disclosures were reported.
Ethical Approval
This research is approved by The George Washington University Committee on Human Research, Institutional Review Board (IRB), FWA00005945. The GW case number is 180804.
Our crawler targeted Weibo’s REST interface at https://s.weibo.com/realtime?.
According to the Regulation on the Administration of Circulation and Vaccination of Vaccines (Supplement 1), China follows the Expanded Program on Immunization (EPI) proposed by the World Health Organization (WHO), which makes nine types of vaccine mandatory. Article 26 of this Regulation states that all new-born babies must register and initiate a vaccination record within one month after birth, whereas article 27 of the Regulation states that all kindergarten and elementary schools must verify students’ vaccination records prior to their enrollment. Article 69 further states that school will face stern punishment for admitting unvaccinated students. Together, these three articles constitute the “no vaccine, no education” policy.
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
References
- 1.Attwell K, Navin MC, Lopalco PL, Jestin C, Reiter S, & Omer SB (2018). Recent vaccine mandates in the United States, Europe and Australia: A comparative study. Vaccine, 36(48), 7377–7384. [DOI] [PubMed] [Google Scholar]
- 2.BBC. (2016). China Shandong illegal vaccine scandal sparks anger. BBC; https://www.bbc.com/news/world-asia-china-35859927. [Google Scholar]
- 3.Betsch C, Brewer NT, Brocard P, Davies P, Gaissmaier W, Haase N, Leask J, Renkewitz F, Renner B, Reyna VF, & others. (2012). Opportunities and challenges of Web 2.0 for vaccination decisions. Vaccine, 30(25), 3727–3733. [DOI] [PubMed] [Google Scholar]
- 4.Blei DM, Ng AY, & Jordan MI (2003). Latent dirichlet allocation. Journal of Machine Learning Research, 3(Jan), 993–1022. [Google Scholar]
- 5.Broniatowski DA, Jamison AM, Qi S, AlKulaib L, Chen T, Benton A, Quinn SC, & Dredze M (2018). Weaponized health communication: Twitter bots and Russian trolls amplify the vaccine debate. American Journal of Public Health, 108(10), 1378–1384. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Cao L, Wang H-Q, Zheng J-S, Yuan P, Cao L-S, Zhang G-M, & others. (2012). National immunization coverage survey in China after integrated more vaccines into EPI since 2008. Chinese Journal of Vaccines and Immunization, 5, 007. [Google Scholar]
- 7.China.org.cn. (2010). Shanxi vaccine scandal. China.Org.Cn; http://www.china.org.cn/china/2010-04/08/content_19769756.htm. [Google Scholar]
- 8.Dixon G, & Clarke C (2013). The effect of falsely balanced reporting of the autism–vaccine controversy on vaccine safety perceptions and behavioral intentions. Health Education Research, 28(2), 352–359. [DOI] [PubMed] [Google Scholar]
- 9.Dredze M, Broniatowski DA, Smith MC, & Hilyard KM (2016). Understanding vaccine refusal: Why we need social media now. American Journal of Preventive Medicine, 50(4), 550–552. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Dubé E, Vivion M, & MacDonald NE (2015). Vaccine hesitancy, vaccine refusal and the anti-vaccine movement: Influence, impact and implications. Expert Review of Vaccines, 14(1), 99–117. [DOI] [PubMed] [Google Scholar]
- 11.Fu K, Chan C, & Chau M (2013). Assessing censorship on microblogs in China: Discriminatory keyword analysis and the real-name registration policy. IEEE Internet Computing, 17(3), 42–50. [Google Scholar]
- 12.Gollust SE, Dempsey AF, Lantz PM, Ubel PA, & Fowler EF (2010). Controversy undermines support for state mandates on the human papillomavirus vaccine. Health Affairs, 29(11), 2041–2046. [DOI] [PubMed] [Google Scholar]
- 13.Hong J, & Wei D (2018, August 7). How Faulty Vaccines Sparked Consumer Panic in China. Bloomberg. https://www.bloomberg.com/news/articles/2018-08-07/how-faulty-vaccines-sparked-consumer-panic-in-china-quicktake
- 14.Kennedy J (2019). Populist politics and vaccine hesitancy in Western Europe: An analysis of national-level data. European Journal of Public Health, 29(3), 512–516. [DOI] [PubMed] [Google Scholar]
- 15.Lama Y, Hu D, Jamison A, Quinn SC, & Broniatowski DA (2019). Characterizing Trends in Human Papillomavirus Vaccine Discourse on Reddit (2007–2015): An Observational Study. JMIR Public Health and Surveillance, 5(1), e12480. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Orr D, Baram-Tsabari A, & Landsman K (2016). Social media as a platform for healthrelated public debates and discussions: The Polio vaccine on Facebook. Israel Journal of Health Policy Research, 5(1), 34. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Radcliffe D (2017). Understanding social media in China. International Journalists’ Network. https://ijnet.org/en/story/understanding-social-media-china. [Google Scholar]
- 18.Tan R (2018). Vaccine scandal gripping China could cause serious problems for the government. https://web.archive.org/web/20180816083403/https://www.washingtonpost.com/news/worldviews/wp/2018/07/26/vaccine-scandal-gripping-china-could-cause-serious-problems-for-the-government/.
- 19.Lancet The. (2018). Vaccine scandal and confidence crisis in China. Lancet (London, England), 392(10145), 360. [DOI] [PubMed] [Google Scholar]
- 20.Xu X, Mao ZM, & Halderman JA (2011). Internet censorship in China: Where does the filtering occur? International Conference on Passive and Active Network Measurement, 133–142. [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.