Weibo-COV: A Large-Scale COVID-19 Social Media Dataset from Weibo

Yong Hu†, Heyan Huang†, Anfan Chen‡, Xian-Ling Mao†

†Beijing Institute of Technology{huyong,hhy63,maoxl}@bit.edu.cn

‡University of Science and Technology of Chinacaf16@ustc.edu.cn

AbstractWith the rapid development of COVID-19,people are asked to maintain “social distance”and “stay at home”. In this scenario, moreand more social interactions move online, es-pecially on social media like Twitter andWeibo. People post tweets to share informa-tion, express opinions and seek help duringthe pandemic, and these tweets on social me-dia are valuable for studies against COVID-19, such as early warning and outbreaks de-tection. Therefore, in this paper, we releasea novel large-scale COVID-19 social mediadataset from Weibo called Weibo-COV1, cov-ering more than 40 million tweets from 1 De-cember 2019 to 30 April 2020. Moreover, thefield information of the dataset is very rich, in-cluding basic tweets information, interactiveinformation, location information and retweetnetwork. We hope this dataset can promotestudies of COVID-19 from multiple perspec-tives and enable better and faster researches tosuppress the spread of this disease.

1 Introduction

At the beginning of writing, COVID-19, an infec-tious disease caused by a coronavirus discovered inDecember 2019, also known as Severe Acute Res-piratory Syndrome Coronavirus 2 (SARS-CoV-2),has infected 4,517,399 individuals globally with adeath toll of 308,515 (Doctor, 2020). Under thecircumstances, the physical aspects of connectionand human communication outside the householdamong people are limited considerably and mainlydepend on digital device like mobile phone or com-puters (Abdul-Mageed et al., 2020). In this kind ofscenario, people will stay at home and spend moretime on the social media communication. The so-cial media plays an important role for people shar-ing information, expressing opinions and seeking

1https://github.com/nghuyong/weibo-public-opinion-datasets

help (Lopez et al., 2020), which makes social me-dia platforms like Weibo, Twitter, Facebook andYoutube a more crucial sources of information dur-ing the pandemic.

In the previous studies, social media was con-sider as a valuable data source for studies againstdisease, like uncovering the dynamics of an emerg-ing outbreak (Zhang and Centola, 2019), predictingthe flu activity and disease surveillance (Jeremyet al., 2009). For example, some studies facilitatebetter influenza surveillance, like early warningand outbreaks detection (Kostkova et al., 2014;De Quincey and Kostkova, 2009), forecast esti-mates of influenza activity (Santillana et al., 2015)and predict the actual number of infected cases(Lampos and Cristianini, 2010; Szomszor et al.,2010). Therefore, it is necessary to make the rel-evant social media datasets freely accessible forbetter public outcomes to facilitate the related stud-ies of COVID-19.

In this paper, we release a novel large-scaleCOVID-19 social media dataset from Weibo, oneof the most popular Chinese social media platform.The dataset is named Weibo-COV and covers morethan 40 million tweets from 1 December 2019 to 30April 2020. Specifically, unlike the conventionalAPI-based data construction methods, which limitlarge-scale data access, we first build a high-qulityWeibo active user pool with 20 million active usersfrom over 250 million users, then collect all activeusers’ tweets during the time period and filter outtweets related to COVID-19 by selected 179 key-words. Moreover, the fields of tweets in the datasetis very rich, including basic tweets information,interactive information, location information andretweet network. We hope this dataset can promotestudies of COVID-19 from multiple perspectivesand enable better and faster researches to suppressthe spread of this disease.

arX

iv:2

005.

0917

4v5

[cs

.SI]

3 J

un 2

020

Figure 1: The construction of Weibo active user pool

2 Data Collection

2.1 Collection Strategy

At present, given specified keywords and a spec-ified period, there are two kinds of methods forconstructing Weibo tweet datasets: (1) Applyingadvanced search API given by Weibo; (2) Travers-ing all Weibo users, collecting all their tweets inthe specified period, and then filtering tweets withspecified keywords.

However, for the first kinds of method, due tothe limitation of the Weibo search API, the result ofone time search contains up to 1000 tweets, mak-ing it difficult to build large-scale datasets. Asfor the second kinds of method, although we couldbuild large-scale datasets with almost no omissions,traversing all billions of Weibo users requires verylong time and large bandwidth resources. In addi-tion, a large number of Weibo users are inactive,and it makes no sense to traverse their homepages,because they may not post any tweets in the speci-fied period.

To alleviate these limitations, we propose a novelmethod to construct Weibo tweet datasets, whichcan build large-scale datasets with high construc-tion efficiency. Specifically, we first build and dy-namically maintain a high-quilty Weibo active userpool (just a small part of all users), and then weonly traverse these users and collect all their tweetswith specified keywords in the specified period.

2.2 Weibo Active User Pool

As shown in Figure 1, based on initial seed usersand continuous expansion through social relation-ship, we first collect more than 250 million Weibousers. Then we define that Weibo active usersshould meet the following 2 characteristics: (1) Thenumber of followers, fans and tweets are all morethan 50; (2) The latest tweet is posted in 30 days.Therefore, we can build and dynamically maintaina Weibo active user pool from all collected Weibousers. Finally, the constructed Weibo active userpool contains 20 million users, accounting for 8%

Table 1: The field description of the dataset

Field Descriptionid the unique identifier of the

tweetcrawl time crawling time of the tweetcreated at creating time of the tweetlike num the number of like at the

crawling timerepost num the number of retweet at the

crawling timecomment num the number of comment at

the crawling timecontent the content of the tweetorigin weibo the id of the origin tweet,

only not empty when thetweet is a retweet one

geo info information of latitude andlongitude, only not emptywhen the tweet contains thelocation information

of the total number of Weibo users.

2.3 COVID-19 Tweets CollectionAccording to the collection strategy described insection 2.1, we set the time period from 00:00 De-cember 1, 2019 (GMT+8, the date of the first di-agnosis) to 23:59 April 30, 2020 (GMT+8), anddesign a total of 179 COVID-19 related keywords.These keywords are comprehensive and rich, cover-ing related terms such as coronavirus and pneumo-nia, as well as specific locations (e.g., “Wuhan”),drugs (e.g, “remdesivir”), preventive measures(e.g., “mask”), experts and doctors (e.g., “ZhongNanshan”), government policy (e.g, “postpone thereopening of school”) and others (see Appendix.1for the complete list).

As a result, based on 20 million Weibo activeuser pool, we first collect a total of 692,792,816tweets posted by these users in the specified period.Subsequently, we filter these tweets by keywordsand finally obtain 40,893,953 tweets. These tweetsconstitute our final dataset.

3 Data Properties

3.1 The Inner Structure of the DatasetAs shown in Table 1, fields of tweetsin the dataset is very rich, covering thebasic information ( id, crawl time,content), interactive information (like num,

Figure 2: The daily distribution of Weibo-COV

Figure 3: Distribution of location information of tweets on April 4, 2020

Table 2: The basic statistics of Weibo-COV

#ALL #GEO #Original40,893,953 1,119,608 8,284,992

repost num, comment num), location in-formation (geo info) and retweet network(origin weibo). Therefore, various aspectsof studies related to infectious diseases can beconducted based on this dataset, such as the impacton people’s daily life, the early characteristics ofthe disease and government anti-epidemic policies.

3.2 Basic Statistic

As shown in Table 2, Weibo-COV contains a totalof 40,893,953 tweets. Among these tweets, thereare 1,119,608 tweets with geographic location in-formation (accounting for 2.7%) and 8,284,992original tweets (accounting for 20.26%).

3.3 Daily DistributionThe distribution of the number of tweets by dayis shown in Figure 2. It can be found that fromDecember 1, 2019 to January 18, 2020, the num-ber of COVID-19 related tweets is very small (lessthan 10K) and may include some noise data. SinceJanuary 19, 2020, the number of COVID-19 re-lated tweets increase rapidly and maintain at least200,000 per day.

Note that the data on April 4, 2020 is particu-larly striking and the number of tweets on that dayexceeds 1.8 million. Because that day was ChineseTomb Sweeping Festival, and a national mourningwas held for the compatriots who died in the epi-demic. People posted or reposted a lot of mourningtweets on Weibo on that day.

3.4 GEO DistributionAs shown in Figure 3, we plot location distribu-tion of tweets with location information on April 4,

Figure 4: Word cloud of tweets in four days and some words are translated in red

2020. It can be seen that the distribution of tweetsis mainly concentrated in China. There are also apart of tweets distributed around the world includ-ing major countries in Asia, Europe, Australia andAmerica. The reason is that with the developmentof economic globalization, more and more Chinesepeople go abroad and more and more foreignersstart to use Weibo.

Therefore, our dataset can study the impact ofthe disease on China in depth and also the impacton the whole world.

3.5 Word Cloud

We select four days of tweets data at differentstages of the epidemic development and draw wordclouds. As shown in Figure 4 (a), in the earlydays, people did not know the characteristics ofthe virus and the government began to take pre-liminary actions (e.g., “unexplained pneumonia”and “health committee”). Later, as shown in Fig-ure 4 (b), people learned that the virus is a newcoronavirus and studied prevention methods andmedicines (e.g., “new coronavirus”, “N95 musk”and “ShuangHuangLian”). Then, as shown in Fig-ure 4 (c), governments took strict isolation rulesand strove to prevent imported cases from abroad(e.g., “isolated at home” and “overseas import”).By the end of April, as shown in Figure 4 (d), thevirus has had many impacts on people’s lives. For-tunately, research on vaccines and medicines hasbeen ongoing and has made effective progress (e.g.,

“Cirque du Soleil in Canada” and “Remdesivir”).Therefore, our dataset runs through the whole

development of COVID-19, and includes impactsof the disease on all aspects of the society.

4 Related Work

Several works have focused on creating socialmedia datasets for enabling COVID-19 research.(Chen et al., 2020), (Lopez et al., 2020) and (Abdul-Mageed et al., 2020) have already released datasetscollected from Twitter. However, these datasets aremainly in English, Chinese tweets are also valu-able and can provide additional supplements forresearches.

Only one dataset proposed by (Gao et al., 2020)includes tweets from Weibo, but their method basedon Weibo advanced search API, so they can notcollect large-scale tweets from Weibo. Comparedwith our dataset, their overall size (less than 200K),time span (from January 20, 2020 to March 24,2020), and number of keywords (only 4 keywords)are all much smaller.

5 Conclusion

In this paper, we release Weibo-COV, a first large-scale COVID-19 tweets dataset from Weibo. Thedataset contains over 40 million tweets coveringfrom 1 December 2019 to 30 April 2020 and eachtweet with rich field information. We hope thisdataset could promote and facilitate related studieson COVID-19.

ReferencesMuhammad Abdul-Mageed, AbdelRahim Elmadany,

Dinesh Pabbi, Kunal Verma, and Rannie Lin. 2020.Mega-cov: A billion-scale dataset of 65 languagesfor covid-19.

Emily Chen, Kristina Lerman, and Emilio Ferrara.2020. Covid-19: The first public coronavirus twit-ter dataset.

Ed De Quincey and Patty Kostkova. 2009. Early warn-ing and outbreak detection using social networkingwebsites: The potential of twitter. In InternationalConference on Electronic Healthcare, pages 21–24.Springer.

Dingxiang Doctor. 2020. Covid-19 global pan-demic real-time reports. https://ncov.dxy.cn/ncovh5/view/pneumonia.

Zhiwei Gao, Shuntaro Yada, Shoko Wakamiya, andEiji Aramaki. 2020. Naist covid: Multilingualcovid-19 twitter and weibo dataset.

Jeremy, Ginsberg, Matthew, H, Mohebbi, Rajan, S, Pa-tel, Lynnette, and Brammer and. 2009. Detectinginfluenza epidemics using search engine query data.Nature.

Patty Kostkova, Martin Szomszor, and ConnieSt. Louis. 2014. # swineflu: The use of twitter asan early warning and risk communication tool in the2009 swine flu pandemic. ACM Transactions onManagement Information Systems (TMIS), 5(2):1–25.

Vasileios Lampos and Nello Cristianini. 2010. Track-ing the flu pandemic by monitoring the social web.In 2010 2nd international workshop on cognitive in-formation processing, pages 411–416. IEEE.

Christian E. Lopez, Malolan Vasu, and Caleb Galle-more. 2020. Understanding the perception ofcovid-19 policies by mining a multilanguage twitterdataset.

Mauricio Santillana, Andre T Nguyen, Mark Dredze,Michael J Paul, Elaine O Nsoesie, and John SBrownstein. 2015. Combining search, social me-dia, and traditional data sources to improve influenzasurveillance. PLoS computational biology, 11(10).

Martin Szomszor, Patty Kostkova, and Ed De Quincey.2010. # swineflu: Twitter predicts swine flu out-break in 2009. In International conference on elec-tronic healthcare, pages 18–26. Springer.

Jingwen Zhang and Damon Centola. 2019. Social net-works and health: new developments in diffusion,online and offline. Annual Review of Sociology,45:91–109.

A Appendices

A.1 Covid-19 Related Keywords

Table 3: The list of selected keywords related to COVID-19

Keywords Translations冠状 CoronavirusCov-19 Cov-19新冠 Coronavirus感染人数 Infected casesN95 N95 Mask大众畜牧野味店 Dazhong wildlife restaurant华南野生市场 South China wild market管轶 Guan Yi武汉病毒所 Wuhan Institute of VirologyCDC Center for Disease Control and Prevention中国疾病预防控制中心 Chinese Center for Disease Control and Prevention疾控中心 Center for Disease Control and Prevention#2019nCoV #2019nCoV双黄连 AND抢购 Shuanghuanglian AND Rush to buy双黄连 AND售磬 Shuanghuanglian AND Sold out武汉卫健委 Wuhan Municipal Health Committee湖北卫健委 Health Commission of Hubei Province#nCoV #nCoVPHEIC PHEIC疫情 Epidemic outbreak火神山 Huoshen Shan hospital雷神山 Leishen Shan hospital钟南山 Zhong NanshanCoronavirus CoronavirusRemdesivir Remdesivir瑞德西韦 Remdesivir感染 AND例 Infected AND cases武汉 AND封城 Wuhan AND Lockdown高福 George Fu Gao王延轶 Wang Yanyi舒红兵 Shu Hongbing协和医院 Xiehe Hospital武汉 AND隔离 Wuhan AND Quarantine李文亮 AND医生 Doctor AND Li Wenliang云监工 Supervising work on cloud武汉仁爱医院 Wuhan Ren’ai Hospital黄冈 AND感染者 Huanggang AND Infected cases孝感 AND感染者 Xiaogan AND Infected cases居家隔离 Isolated at home防护服 Protective Clothing隔离14天 Isolation AND 14 days潜伏期 AND 24天 Incubation period AND 24 days潜伏期 AND 14天 Incubation period AND 14 days国际公共卫生紧急事件 International Public Health Emergencies

Continued on next page

Table 3 – continued from previous pageKeywords Translations方舱医院 AND武汉 FangCang Hospital AND Wuhan一省包一市 one province gives a hand to one Hubei city晋江毒王 Super spreader of COVID-19 in Jinjiang超级传播者 Super spreader湖北 AND王晓东 Hubei AND Wang Xiaodong蒋超良 Jiang Chaoliang李文亮 Li Wenliang千里投毒 Spread Virus from a thousand miles武汉病毒研究 Virology research in Wuhan武汉 AND李医生 Wuhan AND Li Wenliang国家疾控中心 Chinese Center for Disease Control and Prevention武汉 AND疫苗 Wuhan AND Vaccine武汉 AND征用宿舍 Wuhan AND Requisitioned students’ dormitory周佩仪 Zhou Peiyi武汉中心医院 The Central Hospital of Wuhan张晋 AND卫健委 Zhang Jin AND Health Commission张晋 AND卫生将康委员会 Zhang Jin AND Health Commission刘英姿 AND卫健委 Liu Yingzi AND Health Commission刘英姿 AND卫生健康委员会 Liu Yingzi AND Health Commission王贺胜 AND卫健委 Wang Hesheng AND Health Commission王贺胜 AND卫生健康委员会 Wang Hesheng AND Health Commission复工 Enterprise work resuming中小企业 AND困境 Small and medium-sized enterprise AND Dilemma武汉 AND死亡病例 Wuhan AND Death cases武汉 AND感染病例 Wuhan AND Infection cases湖北 AND死亡病例 Hubei AND Death cases湖北 AND感染病例 Hubei AND Infected cases中国 AND死亡病例 China AND Death cases中国 AND感染病例 China AND Infected cases潜伏期 Incubation Period北京 AND病例 Beijing AND Cases天津 AND病例 Tianjin AND Cases河北 AND病例 Hebei AND Cases辽宁 AND病例 Liaoning AND Cases上海 AND病例 Shanghai AND Cases江苏 AND病例 Jiangsu AND Cases浙江 AND病例 Zhejiang AND Cases福建 AND病例 Fujian AND Cases山东 AND病例 Shandong AND Cases广东 AND病例 Guangdong AND Cases海南 AND病例 Hainan AND Cases山西 AND病例 Shanxi AND Cases内蒙古 AND病例 Inner Mongolia AND Cases吉林 AND病例 Jilin AND Cases黑龙江 AND病例 Heilongjiang AND Cases安徽 AND病例 Anhui AND Cases江西 AND病例 Jiangxi AND Cases河南 AND病例 Henan AND Cases

Continued on next page

Table 3 – continued from previous pageKeywords Translations湖北 AND病例 Hubei AND Cases湖南 AND病例 Hunan AND Cases广西 AND病例 Guangxi AND Cases四川 AND病例 Sichuan AND Cases贵州 AND病例 Guizhou AND Cases云南 AND病例 Yunnan AND Cases西藏 AND病例 Tibet AND Cases陕西 AND病例 Shanxi AND Cases甘肃 AND病例 Gansu AND Cases青海 AND病例 Qinghai AND Cases宁夏 AND病例 Ningxia AND Cases新疆 AND病例 Xinjiang AND Cases香港 AND病例 Hong Kong AND Cases澳门 AND病例 Macau AND Cases台湾 AND病例 Taiwan AND CasesECOM Extracorporeal Membrane Oxygenationsars-cov-2 sars-cov-2复学 Resumption of schooling护目镜 Goggles核酸检测 nucleic acid testing (NAT)COVID-19 COVID-192019-nCoV 2019-nCoV疑似 AND病例 Suspicious cases无症状 Asymptomatic Patients累计病例 Cumulative confirmed cases境外输入 imported cases of NCP累计治愈 Cumulative cured cases绥芬河 Sui Fenhe舒兰 Shu Lan健康码 Health QR code出入码 Community Access Code返校 Back to Camp美国 AND例 USA cov-19 AND Cases西班牙 AND例 Spain cov-19 AND Cases新加坡 AND例 Singapore cov-19 AND Cases加拿大 AND例 Canada cov-19 AND Cases英国 AND例 UK cov-19 AND Cases印度 AND例 India cov-19 AND Cases日本 AND例 Japan cov-19 AND Casess韩国 AND例 South Korea cov-19 AND Cases德国 AND例 Germany cov-19 AND Cases法国 AND例 France cov-19 AND Cases意大利 AND例 Italy cov-19 AND Cases新增 AND例 New cov-19 AND Cases人工膜肺 Extracorporeal Membrane Oxygenation双盲测试 Double Blind Test疫苗 Vaccine小区出入证 Community Entry card

Continued on next page

Table 3 – continued from previous pageKeywords Translations战疫 Anti-COVID-19抗疫 Anti-COVID-19湖北卫健委 AND免职 Health commission of Hubei Province AND Remove from the position发热患者 Fever patients延迟开学 Postpone the reopening of school开学时间 AND不得早于 The start time of school AND Not earlier than累计死亡数 Cumulative deaths疑似病例 Suspicious cases入户排查 Household troubleshoot武汉 AND肺炎 Wuhan AND Pneumonia新型肺炎 Novel Pneumonia不明原因肺炎 Pneumonia of unknown cause野味肺炎 Wildlife pneumonia出门 AND戴口罩 Going out AND Wear mask3M AND口罩 N95 AND MaskKN95 AND口罩 3M AND Mask新肺炎 Novel Pneumonia#2019nCoV #2019nCoV新型肺炎 AND死亡 Novel Pneumonia AND Death新型肺炎 AND感染 Novel Pneumonia Infection武汉 AND肺炎 AND谣言 Wuhan AND Pneumonia AND Rumors8名散布武汉肺炎谣言 Eight people AND Spread rumors of Wuhan pneumonia黄冈 AND新肺炎 Huanggang AND Novel Pneumonia孝感 AND新肺炎 Xiaogan AND Novel Pneumonia居家隔离 Isolated at home武汉中心医院 AND新型肺炎 The Central Hospital of Wuhan AND Novel Pneumonia武汉肺炎 Wuhan Pneumonia企业复工 Enterprise work resuming囤积口罩 Hoarding mask零号病人 Zero Patient黄燕玲 Huang Yanling病毒源头 Oringin of Cov-19电子烟肺炎 AND新型冠状 E-cigarette Pneumonia AND Coronavirus病毒战 Virus War病毒 AND实验室泄露 Virus AND laboratory leakage比尔盖茨 AND疫苗牟利 Bill Gates AND Vaccine for profit美国细菌实验室 US Army Bacterial Laboratory确诊 Confired Infencted COV-19 casespandemic pandemic