In December 2019, an outbreak of COVID-19, caused by infection with SARS-CoV-2, was found and publicly reported in China and subsequently became a global pandemic (1–5). As of 31 May 2020, more than 5.9 million cases were confirmed in most countries and territories worldwide, with more than 367,000 fatalities (6), seriously threatening the public's lives and health and menacing economic stability and social security of nations.

Public health emergencies are often accompanied by the Internet public opinion crisis characterized by frequent rumors, especially when official authorities' information is delayed or lacking (7). Since the advent of Web 2.0 technology, Internet social media platforms, such as Weibo (similar to Twitter), have gradually replaced traditional media as the dominant platforms for the Chinese public to express their opinions and participate in social affairs. As of the last quarter of 2020, Sina Weibo had about 521 million active users per month, increasing 10 million over the same period last year (8). Due to social media's easy availability and convenience, information spreads more rapidly and widely through these platforms than its traditional counterparts (9, 10). Moreover, the resulting mass of user-provided content promotes vast recruitment of people around shared interests, worldviews, and narratives, thus influencing the evolution of public opinion (11) and further enabling rumors to thrive. In 2013, the World Economic Forum described web-based rumors as ‘digital wildfire' and accentuated the risks they pose to modern society (12). “Drinking ardent spirits can prevent SARS-CoV-2 infection” was a typical example of a rumor spread widely during the outbreak of COVID-19 in China. Since the epidemic outbreak, social media users rushed to the Internet to seek methods of preventing and treating COVID-19. Due to strong concerns about their own lives and the lack of awareness of the disease, many microbloggers distorted the causal relationship between COVID-19 and drinking, and these blogs became very pervasive (13). These false messages were widely discussed on the Internet at the time, causing great chaos and scares.

Compared with the severe acute respiratory syndrome (SARS) outbreak 17 years ago, the COVID-19 outbreak has sparked more rumormongering. Rumors such as ‘Dual yellow oral liquid inhibited novel coronavirus,' ‘The number of confirmed cases of COVID-19 and deaths in a county,' and ‘Some places have been blockaded or the supermarkets have been closed down' sparked panic among the public, triggering people to snap up supplies and posing a severe challenge to the governance of Internet public opinion in the context of the pandemic. Therefore, it is significant to study the characteristics of rumor generation and propagation on the network platform and propose appropriate approaches to suppress rumors effectively.

The following paper is organized as follows: Section Related Work introduces related works of experts in correlative areas and states research gaps. Section Data and Methods provides an introduction to the data collection, data preprocessing, and analysis methodology. In SectionResults and Discussion, the main results are presented and discussed. Section 5 and Section Strengths and Limitations introduce the strengths and limitations of this study, respectively. Finally, a summary of this work and some meaningful conclusions are presented in Section Conclusion.

False information is generally referred to as information pollution. According to the heterogeneity, Meel and Vishwakarma (14) divided false information into several categories: rumor, fake news, misinformation, disinformation, etc. There are different definitions of rumor in pieces of literature. Gist (15) proposed that rumors have three characteristics: word of mouth, informative, and expressive. With the development of information technology, the transmission mode of rumors has changed radically, and they are no longer passed by word of mouth, but they are more prone to distortion of information. The informative and expressive qualities of rumors, however, have not changed. The informative nature of rumors means that they are intended to convey information about particular people, happening, or conditions. Rumors express and satisfy the emotional needs of the group. Based on this, this paper defines rumors as statements that are fabricated and publicized remarks through specific means without a corresponding factual basis.

It is necessary to carry out a series of research on rumors for rapid damage control. At present, there are four important perspectives of rumor analysis: source detection (16–18), propagation dynamics (19, 20), fake information detection (21–23), containment, and intervention (24, 25). A widely spread rumor or fake news can lead to reputation ruin (26), political consequences (27), and economic loss (28). Numerous scholars are dedicated to revealing the inherent driving mechanisms of rumors and putting forward effective strategies to prevent and control rumors (29–32). The in-depth tracing of misinformation and disinformation across social networks is a complicated process. Bruns et al. (33) drew on a mixture of quantitative and qualitative methods and showed the dynamics of the rumor to uncover the main driving forces and inflection points of COVID/5G conspiracy theories. Islam et al. (34) followed and examined COVID-19–related rumors on various online platforms and further explored their impacts on public health.

When an emergency occurs, many factors drive the generation and transmission of rumors, such as psychology (35), the heterogeneity of individual behaviors (36), opinion leaders (37), etc. There is a complex network relationship between rumors and driving factors. Wang et al. (38) discovered that network topology is an important factor affecting the spread of rumors. Dong et al. (39) found that the angrier the public feel, the more rumors there will likely be. Hui et al. (40) drew an interesting conclusion that the education level of the crowd is an essential factor affecting the final scale of rumor propagation.

There are subtle differences between the rumor generation and dissemination mechanisms. Most studies on rumor influencing factors have been carried out in the propagation mechanism, and plenty of models and methods have been implemented to simulate the spreading process of rumors, such as epidemic models (41–43), forest fire model (44), and explosive explosion principle (45). Some scholars also use questionnaire methods to study the influencing factors of rumors. Sun et al. (46) investigated 556 Chinese people through online questionnaires and obtained the factors influencing middle-aged and older adults regarding the re-spreading of rumors about COVID-19. Experiments showed that network properties profoundly affect the rumor diffusion process, and the complex structure of social networks can be modeled using different graphical formats. Cheng et al. (47) utilized the stochastic epidemic model to analyze rumor propagation on the online social site BlogCatalog by formalizing an undirected graph G (V, E) dataset, which contains 10,312 nodes and 333,983 edges. Wood (48) analyzed the content and social network of 25,162 original tweets about the Zika virus conspiracy theory, pointing out that conspiracy theories spread through a more decentralized network relative to debunking messages. However, the behaviors of rumor makers are the root of curbing rumors and are an effective way to block rumors. Few studies have investigated the mechanism of rumor generation systematically. Zhu and Liu (49) put forward the disinformation behavior evolution model of the rumor maker by the system dynamics, and concluded that the effect of personal effects on rumor tendency is the most significant, followed by social effects, while the government effects have only a subtle effect on the rumor tendency. In summary, the mechanics of online rumors is very complicated and dynamic, making it difficult to prevent a rumor from generating and propagating throughout the Internet. A great deal of previous research on rumors has focused on the spread of rumors. There has been no detailed investigation regarding the rumor generation mechanism.

On the basis of extant literature, it can be argued that the category of false information has been divided (14, 50). Many different factors are available for classifying rumors, such as type, scope, and characteristics. Jaeger et al. (51) classified rumors into believable and unbelievable based on their credibility. Zubiaga et al. (22) had interestingly split rumors into two main categories: long-standing rumors and breaking news rumors. From the perspective of psychology, Knapp (52) categorized rumors into “pipe-dream” rumors, “bogy” rumors, and “wedge-driving” rumors. Therefore, the classification of rumors is not yet uniform. Only a handful of studies have classified rumors during COVID-19 (53). This leads to the first research question (RQ),

As the country was first hit by the pandemic, China experienced a complex social environment during the severe anti-pandemic period. Exploring the influencing factors and trends of related rumors has positive significance for the managers to maintain social stability. Zhu et al. (54) analyzed social media topics and emotional change characteristics during COVID-19 from spatiotemporal perspectives. Chen et al. (53) divided the COVID-19 outbreak into five periods according to the key events and disease epidemic and plotted the trends of the epidemic and the focus of the public at different stages. Motivated by these studies, we incorporate the trend of epidemic and essential news into rumor studies. Thus, the follow-up research question is,

However, emergencies are not homogenous. For instance, the factors and variables that have influenced the rumors related to the Middle East respiratory syndrome outbreak in 2015 and the Zika virus in 2015-16 are different (48, 55). Besides, a lack of research specifies the differences between the causal factors of different types of rumors. More importantly, limited studies have attempted to distinguish between rumor making and rumor spreading. From this, we get the third research question,

To answer this, this study identified 1,537 rumor-related posts about COVID-19 on Weibo, one of China's most popular social media platforms. We explored the classification and characteristics of rumors. The development trend of emergencies and hot events were also taken into account to investigate the spatiotemporal characteristics of rumors. Besides, rumor-driving factors were extracted and discussed.

This section presents the data and methods, including data collection, data processing, and the three main research methods. In the data processing section, we explore RQ1.

In this section, the time evolution characteristics, the spatial and temporal correlation characteristics, and the network characteristics of the rumors are analyzed and mined (RQ2 and RQ3).

Studying and dealing with Internet rumors is a complex scientific issue, and this paper provides new insights into the existing literature. To the authors' best knowledge, this is the first study that attempts to explore the mechanisms of rumors in depth from various perspectives.

Although this study has provided exciting findings for theory and practice, some problems are still worth studying and solving. This paper takes Sina Weibo rumors during the COVID-19 outbreak as a sample, but the analysis of rumors needs to consider different emergency characteristics and social networking platforms. It is better to introduce rumors from different emergencies and social platforms for further analysis in future research.

This is an in-depth and detailed study on rumors during the outbreak of novel coronavirus pneumonia. It can be concluded from the correlation analysis that the number of confirmed cases during the pandemic has a positive effect on the number of rumors. In detail, there was a change from a positive correlation to a negative correlation between the daily number of rumors and the cumulative confirmed cases, which indicated that the epidemic trend had generally become steady. The second significant finding was that the risk level of Creating Panic-type rumors was the highest and presented a sharp increase trend in the early stage of the epidemic. The last worth-highlighting point was the analysis of the rumor cause network, and we discovered that panic psychology and the lag of information disclosure were the crucial reasons for generating and propagating all kinds of rumors. In addition, the uncertainty of epidemic development played a vital role in the generation of rumors. The above-mentioned findings demonstrate that it is necessary to take timely and effective epidemic prevention and control measures and disclose information timely, which are conducive to reducing the public's negative mentality and controlling rumors. Besides, the confusing nature of rumors and the public's low scientific literacy were essential driving factors of rumor diffusion. The government needs to enhance the popularization of relevant science knowledge accordingly. The findings are conducive to authorities adopting targeted measures and further improving the level of social management.

The original contributions presented in the study are included in the article/supplementary material, further inquiries can be directed to the corresponding author/s.

WL: funding acquisition, validation, conceptualization, and writing—original draft. WZ and WL: formal analysis, methodology, and writing—review and editing. BG: writing—review and editing, funding acquisition, validation, and supervision. YH and HF: validation and writing—review and editing. All authors contributed to the article and approved the submitted version.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.