Abstract
Social media has become a mainstream channel of communication during the COVID‐19 pandemic. While some studies have been developed on investigating public opinion on social media data regarding COVID‐19 pandemic, there is no study analyzing anti‐quarantine comments on social media. This study has collected and analyzed near 80,000 tweets to understand anti‐quarantine social comments. Using text mining, we found 11 topics representing different issues such as comparing COVID‐19 and flu and health side effects of quarantine. We believe that this study shines a light on public opinion of people who are against quarantine.
Keywords: COVID‐19, quarantine, social media, twitter, text mining
1. INTRODUCTION
As of June 4, 2020, the number of positive cases was more than 6.6 million globally, with over 390,000 deaths. 1 The outbreak has posed significant threats to international health and the economy. In the absence of treatment for this virus, US states imposed a quarantine period. In addition, the COVID‐19 pandemic caused the largest global recession in history, 2 and the US unemployment rate reached 14.7% in April 2020. 3
Social media has become a mainstream channel of communication (Turner‐McGrievy, et al., 2020). For example, 72% of U.S. adults use at least one social media site in 2019. 4 In the last decade, social media platforms such as Twitter have grown in popularity. Social media has received a great deal of academic interest to investigate public opinion on different issues such as politics and health (Karami, Lundy, et al., 2020). Some studies, such as (Abd‐Alrazaq et al., 2020), have been developed on analyzing public opinion on social media data during the COVID‐19 pandemic. While these studies have provided valuable insights, they did not consider anti‐quarantine comments on social media. This study characterizes tweets containing anti‐quarantine hashtags with text mining to present a new perspective on public opinion during the COVID‐19 pandemic.
2. METHODOLOGY
2.1. Data collection and preprocessing
We explored Twitter search to identify hashtags representing anti‐quarantine semantic. We found six hashtags including #AntiLockdown, #AntiQuarantine, #ReOpenAmerica, #EndtheLockdown, #EndtheLockdown, and #ReOpenAmericaNow. Then, we utilized a data provider service (Brandwatch) to collect tweets containing at least one of the six hashtags. Next, we removed URLs, hashtags, username, duplicate tweets and tweets with less than five words. This process has provided 79,738 tweets.
2.2. Topic discovery
In order to disclose hidden semantic layer in our data, we utilized topic modeling. This technique assigns words semantically related to each other into a cluster, called topic. Different topic models have been developed using Latent Dirichlet Allocation model (LDA) (Blei et al., 2003). LDA is an efficient and effective topic model that has been applied on different social media applications such as analyzing exercise patterns (Karami and Shaw, 2019) and mobile work (Hemsley et al., 2020) on Twitter. The outputs of LDA for n documents (tweets), m words, and t topics, are two matrices (Karami, Ghasemi, et al., 2019). The first one is the probability of each word in each topic or P(W i |T k ), and the second one is the probability of each of topic in each document or P(T k |D j ) (Figure 1).
FIGURE 1.
Outputs of LDA
The P(W i |T k ) matrix output represents the probability of each word belonging to a specific topic, whereas the P(T k |D j ) matrix represents the probability of each topic belonging to a specific document (Karami, White, et al., 2020). The top words in each topic based on the descending order P(W i |T k ) represent a topic. For example, the top words in topic 4 in Table 1 are people, home, stay, sick, masks, live, care, risk, wear, and safe, representing a theme related to wearing masks. We used a pre‐processing step to find the optimal number of topics based on the level of consistency and coherence with the associated topic using the C_V method developed in the gensim Python package. 5 The estimation process offered the number of topics at 11. Next, we applied Mallet, a Java implementation of LDA, at 11 topics and 4,000 interactions on our corpus.
TABLE 1.
Topics of anti‐quarantine tweets
| ID | Label | Words per topic |
|---|---|---|
| T1 | Unconstitutional order | Rights governors freedom government state people orders constitution free unconstitutional |
| T2 | Health issues | Economy end lockdown virus people lives health worse time risk |
| T3 | Small business | Open essential social businesses time closed politicians small stores corrupt |
| T4 | Masks | People home stay sick masks live care risk wear safe |
| T5 | Blocking small business support | Businesses money pay families reason government political save democrat bills |
| T6 | Open businesses | Back America work time country reopen business save open jobs |
| T7 | Opening plan | Reopening reopen states plan time testing weeks economy start trump |
| T8 | Wrong models and fake news | News media wrong science data models daily truth stop fake |
| T9 | Protests | State protest governor reopen today California rally patriots Michigan join |
| T10 | Political consequences | Trump president hell love thing democrats party white hoax election |
| T11 | Flu vs COVID‐19 | Deaths people covid flu cases million virus rate numbers population |
3. TOPIC ANALYSIS
We then moved to a qualitative coding process to inductively interpret topics. Our coding was based on reviewing the top related words within each topic using P(W|T) and the top related tweets within each topic using P(T|D) (Karami, Swan, et al., 2019).
4. RESULTS
We found anti‐quarantine tweets discussed in 11 topics, as seen in Table 1. The first topic showed discussions on the order of governors to close businesses and impose self‐quarantine. These tweets argued that the order was unconstitutional because it was against people's liberty and freedom. The second topic represented arguments around the physical and mental health issues due to quarantine. The third topic focused on this question: “Why were big businesses (e.g., Walmart) open while small businesses were closed?”
The fourth topic was about the uselessness of wearing a mask for safety. The fifth topics considered blocking a republican bill that added small business loans by senate democrats. The sixth and seventh topics showed requests for open businesses and the Opening Up America, respectively. The next topic illustrated debates on fake news and wrong models on predicting COVID‐19 related issues. The ninth topic was about protests again quarantine in different US states such as California and Michigan. The next topic was discussion on political consequences of the COVID‐19 quarantine in elections. The last topic compared flu and COVID‐19 based on different statistics such as the number of deaths. Table 2 also provides an example tweet for each of topics.
TABLE 2.
Tweet examples
| Topic | Tweet example | Topic | Tweet example |
|---|---|---|---|
| Unconstitutional order | Please support Elon Musk @elonmusk as he battles against Leviathan (the government) for our rights and liberty as individuals and private businesses. A true revolutionary is one who defies the State at his potential peril. #business #tesla #covid #california #ReopenAmericaNow | Open businesses | America MUST get back to work. Click here to sign the petition to save jobs, small business, and America as we know it! #reopenamerica #reopenamericanbusiness |
| Health issues | Absolutely & we MUST also remember those that have lost their lives during this dangerous lockdown due to addiction, suicide & other reasons, all due to the extended unnecessary lockdown. The lockdown has caused immense harm to MANY ! #ReOpenAmerica | Opening plan | President Trump unveiled ‘Opening Up America' plan, aims for May 1. The guidelines were developed by #coronavirus task force members, Drs. Birx & Fauci in coordination with CDC & Prevention chief Robert Redfield. #ReopenAmerica |
| Small Business | Ian Smith, owner of Atilis Gym, is right “enough is enough”. Small business are just as “essential” as big box stores like Wal‐Mart. If big box stores servicing over a 100 customers can take the necessary precautions to remain open so can small businesses. #ReopenAmericaNow | Wrong models and fake news | An analysis from Bot Sentinel, a bot tracking platform, found that bots and trolls have been stoking sentiments online that have fueled the protests, using hashtags like #ReopenAmericaNow and #StopTheMadness. |
| Masks | Absolutely & we MUST also remember those that have lost their lives during this dangerous lockdown due to addiction, suicide & other reasons, all due to the extended unnecessary lockdown. The lockdown has caused immense harm to MANY ! #ReOpenAmerica | Protests | There are protests planned in California, Oregon, Washington, Arizona, New Mexico, Colorado, Kansas, Minnesota, Iowa, Missouri, Illinois, Wisconsin, Indiana, Michigan, Kentucky, Tennessee, Ohio, Pennsylvania, N.Carolina, Delaware and Maine. #ReopenAmerica |
| Blocking small business support | You're not making lists of US taxpayers. | Political consequences | Trump now losing ground in a key swing state. Remember, even Winston Churchill got crushed right after he won World War Two! Want to reelect Trump? #ReopenAmerica |
| #ReopenAmericaNow! Democrats withheld critical funding for the people who pay their salaries; who are struggling to feed their families, pay their bills, save their businesses. You did it for political gain, without missing a paycheck. | |||
| Flu vs COVID‐19 | @realDonaldTrump TESTS are not even required to confirm deaths as Coronavirus. Total deaths for each week in March as compared to FOUR prior years are DOWN. Many flu & pneumonia deaths are coded falsely as coronavirus. That is the only logical conclusion. #ReopenAmericanow |
5. DISCUSSION AND CONCLUSION
This study investigated anti‐quarantine tweets to understand hidden topics Twitter with a cost and time effective approach. Our findings show the anti‐quarantine tweets were about different issues related to the COVID‐19 quarantine. Our research shines a light on US public opinion regarding anti‐quarantine efforts.
While this study provided a new perspective in public opinion mining, we did not study non‐English tweets and this paper was limited to six hashtags. Future work could investigate non‐English tweets and consider more hashtags.
ACKNOWLEDGEMENT
This work is partially supported by the College of Information and Communication Internal Collaboration Grant and the Big Data Health Science Center (BDHSC) at the University of South Carolina. All findings in this paper do not necessarily reflect the views of the funding agency.
Karami A, Anderson M. Social media and COVID‐19: Characterizing anti‐quarantine comments on Twitter . Proc Assoc Inf Sci Technol. 2020;57:e349 10.1002/pra2.349
Endnotes
REFERENCES
- Abd‐Alrazaq, A. , Alhuwail, D. , Househ, M. , Hamdi, M. , & Shah, Z. (2020). Top concerns of tweeters during the COVID‐19 pandemic: Infoveillance study. Journal of Medical Internet Research, 22(4), e19016. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Blei, D. M. , Ng, A. Y. , & Jordan, M. I. (2003). Latent dirichlet allocation. Journal of Machine Learning Research, 3(Jan), 993–1022. [Google Scholar]
- Hemsley, J. , Erickson, I. , Jarrahi, M. H. , & Karami, A. (2020). Digital nomads, coworking, and other expressions of mobile work on Twitter. First Monday, 25. [Google Scholar]
- Karami, A. , Lundy, M. , Webb, F. , & Dwivedi, Y. K. (2020). Twitter and research: A systematic literature review through text mining. IEEE Access, 8, 67698–67717. [Google Scholar]
- Karami, A. , White, C. N. , Ford, K. , Swan, S. , & Spinel, M. Y. (2020b). Unwanted advances in higher education: Uncovering sexual harassment experiences in academia with text mining. Information Processing & Management, 57(2), 102167. [Google Scholar]
- Karami, A. , Ghasemi, M. , Sen, S. , Moraes, M. F. , & Shah, V. (2019). Exploring diseases and syndromes in neurology case reports from 1955 to 2017 with text mining. Computers in Biology and Medicine, 109, 322–332. [DOI] [PubMed] [Google Scholar]
- Karami, A. , & Shaw, G. (2019). An exploratory study of (#) exercise in the Twittersphere. iConference 2019 Proceedings.
- Karami, A. , Swan, S. C. , White, C. N. , & Ford, K. (2019b). Hidden in plain sight for too long: Using text mining techniques to shine a light on workplace sexism and sexual harassment. Psychology of Violence. [Google Scholar]
- Turner‐McGrievy, G. , Karami, A. , Monroe, C. , & Brandt, H. M. (2020). Dietary pattern recognition on Twitter: A case example of before, during, and after four natural disasters. Natural Hazards, 103, 1035–1049. [Google Scholar]