The Chinese online community reacted rapidly to news about infectious disease outbreaks both within and beyond China, as shown in our study. This paper is the first to document this online response using Weibo and to compare the reaction to the MERS-CoV outbreak in 2012 with the reaction to the human infections of avian influenza A(H7N9) in 2013. We found that the reaction to the H7N9 outbreak in 2013 was about two orders of magnitude stronger than the one to the MERS-CoV outbreak in 2012. The results confirmed our hypothesis that the Chinese online community reacted more strongly to an outbreak that was in China than one outside China.
The reaction in the Chinese online community exploded within the first five days of the first case report of three human cases (two in Shanghai and one in Anhui) of avian influenza A(H7N9) [22]. Within these five days, more cases were identified in Shanghai and in two neighboring provinces of Jiangsu and Zhejiang. However, attention soon declined rapidly. It declined until April 13, 2013, when the Chinese government announced that a child was found H7N9-positive in Beijing, the capital of China. This piece of news triggered a second explosion of online discussion via Weibo on that day. Attention then declined rapidly again (Figure 2).
Keywords that were sensitive and specific to the signals were identified. Keywords like "H7N9" and "冠状病毒" (Coronavirus) were highly sensitive and specific. Keywords like "禽流感" (avian flu) and SARS, while less specific, remained sensitive enough to detect the signals.
While the keyword "非典" (Feidian, shortened for atypical pneumonia) was not sensitive to the news of MERS-CoV on September 23, 2012 (Figure 3b), we would like to highlight its significance in the lexicon of the current Chinese online community as one of its most frequently used term for SARS in online discussion. As a keyword, "非典" (Feidian) was sensitive to rumors of SARS in the city of Baoding, China, on February 19, 2012. The rumors were later rejected by the Chinese authorities on February 26, 2012 when the possibility of SARS infection among feverish hospitalized patients in a hospital in Baoding was excluded (Figure 3b) [27]. This keyword, however, also led to a "false positive". On July 21, 2012, there was a severe flood in Beijing, resulting in dozens of deaths. The Chinese online community complained about the Beijing municipal government's disaster management. The government reacted by holding a press conference on July 24, saying that they had learned the lessons of SARS in 2003 and did not conceal the true death toll [28]. This incident also led to a peak in posts with the keyword "非典" (Feidian) (Figure 3b). On January 30, 2013, in a telephone interview with the China Central Television, Prof. ZHONG Nan-Shan, a well-respected medical researcher with a reputation as a leader in fighting against SARS in 2003 in China, mentioned that air pollution in China was more dreadful than "非典" (Feidian) because no one could escape from it [29]. His quote from the interview also led to a peak of Weibo posts with the keyword "非典" (Feidian) (Figure 4).
The observation that Weibo posts with the keywords "非典" (Feidian) and SARS rose to 3131.9 and 1485.4 per million on April 3, 2013 (Figure 4) was consistent with a similar observation in web search query data from Google Trends ([30]; search terms: SARS; "非典"; time range: 2013; Location: China; accessed on October 5, 2013), in which a peak was observed during the week of March 31, 2013. Given China’s SARS experience in 2003, the Chinese online community’s reaction is not surprising. Our observations show that the Chinese online community discussed SARS in the first week after the first report of the H7N9 outbreak with an order of magnitude higher frequency than that in the first week after the first report of the MERS-CoV outbreak. These results again confirmed our hypothesis that the Chinese online community reacted more strongly to an outbreak that happened in China than one outside China.
Drawing on the social amplification of risk model [31], public risk perception is shaped by a process of interplays between psychological, cultural, social, and institutional factors that may result in amplifying or attenuating the public attention to risk. Mass communication is among the list of factors. Public health officials have long recognised the role of the mass media in disseminating risk and emergency information before, during, and after a catastrophe [32]. The World Health Organization establishes guidelines for “effective media communication”, through which the authorities are able to disseminate information to the public [33]. Communication during crisis was traditionally understood to be a one-way and top-down process, in which the public are assumed to be “deficient” in knowledge, while the scientists, public health experts, and emergency managers, are “sufficient” [34]. But this presumption was profoundly challenged by the emergence of social media. For instance, Leung and Nicoll argued that the 2009 H1N1 pandemic was the first pandemic in which social media “challenged conventional public health communication” [35]. In China, online messages were published ahead of the official statement in the 2008 Sichuan Earthquake [36]. Social media enabled people under crisis to share information and experience and to seek message credibility and confirmation via multiple media platforms and social networks [34]. Our study demonstrated that official data released by health authorities, whether in Beijing or Geneva, received strong reactions in the Chinese online community. With such knowledge, social media should be incorporated in the best practices for risk and crisis communication [37]. Social media data can also provide health authorities, researchers and the media a quantifiable measure of public attention towards a particular disease outbreak [11].
Social media, in addition to being a tool to release and track official outbreak information [38], offers a new opportunity for public health practitioners to understand social and behavioral barriers to infection control, to identify misinformation and emerging rumors [39], and to better understand the sentiments and risk perception associated with outbreaks and preventive and control measures [13]. In turn, these will help facilitate better health communication between public health agencies and the society at large, as well as among citizens themselves.
With our Weibo data, there are at least two potential directions for future research. First, we can study how information about a given disease spread across the social network as represented by Weibo. Kwak et al. [40] identified a non-power-law follower distribution, a short effective diameter and low reciprocity in Twitter follower-following topology, which was different from most human social networks. Over 85% of the top trending topics on Twitter are headline news or persistent news. Once retweeted, a tweet would reach an average of 1,000 users regardless of original tweet’s number of followers [40]. However, a previous study has found that Chinese Weibo exhibits a distinct pattern of information dissemination [41]. For example, the network connections between Chinese microbloggers are markedly hierarchical than those between Twitter users, i.e. Chinese users tend to follow those at a higher or similar social level [42]; majority of Weibo posts are indeed re-posts that are originated from a small percentage of original messages [24]. It will be very interesting if further research can shed light upon how information sharing over Weibo can affect human response to the diseases off-line.
Second, content analysis of Weibo posts will enable us to analyze human attitudes or reactions toward health hazard [43]. The research can be extended to investigate anxiety or fear towards the infectious diseases themselves and towards the outbreak information transmitted via the Weibo social network. Similar research on influenza has been conducted using Twitter data [12, 14]. Data mining methods, like topic models [44], may be attempted.
There are a few limitations to our study. The sampled microbloggers in our study were limited to those who have more than 10,000 followers. Despite the fact that these microbloggers are more likely to be authentic users rather than spam accounts, the samples constitute less than 0.1% of the overall microblogger population [23]. However, a random sampling study finds that Weibo content contribution is unevenly distributed among users [23]. Over half of Sina Weibo subscribers have never posted, whereas about 5% of Weibo users contributed more than 80% of the original posts [23]. Hence, the sampled microbloggers in our study were the most influential microbloggers who contributed a majority of Weibo posts and drew the most attention in terms of the number of reposts and comments [23]. Therefore, for the purpose of this study, this group of high-follower-count microbloggers should be deemed fairly representative of the public attention towards the MERS-CoV and H7N9 outbreaks. But the reader should note that the findings of our study might not be generalizable to the samples collected by other sampling strategies. The operational parameters of sampling were not determined to optimize collection of data specific to a given disease. Future research is warranted to reconfirm the research findings by using a research design that is customized for specific epidemiologic research purposes.