Abstract
We randomly extracted publicly available Tweets mentioning COVID-19 related terms (n=2,558,474 Tweets) from Tweet corpora collected daily using an API from Jan 21st to May 3rd, 2020. We applied a clustering algorithm to publicly available Tweets authored by African Americans (n=1,763) to detect topics and sentiment applying natural language processing (NLP). We visualized fifteen topics (four themes) using network diagrams (Newman modularity 0.74). Compared to the COVID-19 related Tweets authored by others, positive sentiments, cohesively encouraging online discussions (e.g., Black strong 27.1%, growing up Blacks 22.8%, support Black business 17.0%, how to build resilience 7.8%), and COVID −19 prevention behaviors (e.g., masks 4.7%, encouraging social distancing 9.4%) were uniquely observed in African American Twitter communities. Application of topic modeling to streaming social media Twitter provides far-reaching research insights into the African American lived experience of COVID-19 and future virtual based interventions to address racial health disparities in COVID-19 experiences.
Keywords: social media, pandemic, virtual intervention, health disparities
1. Introduction
To date, a total of 3,525,116 confirmed COVID-19 cases and 243,540 deaths have been reported worldwide. While there are several ongoing clinical trials, there are no medicines that have been shown to prevent or cure the disease. Notably, African Americans compared to non-Hispanic Whites are disproportionately more likely to die due to Covid-19 (e.g., 70% of COVID-19 deaths in Chicago or Louisiana) [1].
Topic modeling applies statistical analyses to counts of words or word groups that co-occur within text documents in order to reveal latent word groupings with potential semantic significance across a collection of documents. Compared to manual reading and keyword/key phrase extraction algorithms (TF-IDF, TextRank), topic modeling offers the ability to find semantically meaningful groupings of multiple words across very large document collections, even in cases when all of the words do not appear together within any single document or section [2]. Topic modeling is typically applied in health science to facilitate meta-analyses of large collections of research publications, or for generating high-level summaries of discussions (e.g., on social media) among groups of interest. The purpose of the study was to apply topic modeling techniques to Tweets as the foundation to inform designs of culturally sensitive interventions targeting African Americans on virtual platforms for COVID-19.
2. Methods
We applied topic modeling and sentiment analysis for content mining of publicly available Tweet corpus mentioning COVID-19 (keywords: covid19, Covid, covid-19, COVID, corona, coronavirus). We used NCapture, ORA and a Terremoto High Performance Computing Cluster (https://cuit.columbia.edu/shared-research-computing-facility). We randomly extracted publicly available Tweets mentioning COVID-19 related terms (n=2,558,474 Tweets) from Tweet corpora collected daily using an API (https://osf.io/qruf3/?view_only=08302960d9f4486083e82971a1eda125) from Jan 21st to May 3rd, 2020. First, 327 XML files were merged using Perl regular expression commands and shell scripting on the Terremoto cluster. Second, we randomly extracted Covid-19 related Tweets using publicly open African American Twitter community hashtags (#blacktwitter, #staywoke, #blacklivesmatter) [3]. Third, we applied NLP and a clustering Newman algorithm to the publicly available Tweet corpus. The relative frequency of 55,035 n-grams from both corpora were compared using Chi-square in order to identify n-grams uniquely occurring in Tweets authored by public African American Twitter communities. Next, we visualized 15 topics using a network diagram (Newman modularity 0.74). Lastly, experts in African American culture studies, public health, nursing and behavioral science performed qualitative thematic analysis on the detected topics. The larger study was approved by the Institutional Review Board (IRB).
3. Results
Positive sentiments, cohesively encouraging online discussions (e.g., Black strong 27.1%, growing up black 22.8%, support black business 17.0%, how to build resilience 7.8%), and COVID-19 prevention behaviors (e.g., masks 4.7%, encouraging social distancing 9.4%) were uniquely observed in Tweet corpus authored by African Americans. Top 50 unique n-grams in COVID-19 related Tweets authored by African Americans and their frequencies are summarized in Table 1 (χ2 p-value < 2.2e-16).
Table 1.
Top unique n-grams in COVID-19 related public Tweets within African American Communities
| n gram | frequency | n gram | frequency |
|---|---|---|---|
| African Americans | 40.72% | stylist | 6.26% |
| xcellence | 32.66% | Monday motivation | 6.04% |
| black strong | 27.07% | woo young (k-pop singer) | 6.04% |
| growing up black | 22.82% | calling | 5.82% |
| support black business | 17.00% | quarantine | 5.82% |
| warriors | 16.11% | hospitals | 5.59% |
| died (deaths, dying) | 13.20% | mortalities | 5.37% |
| doctors | 12.30% | patients | 5.15% |
| lock down hustle | 11.63% | wards | 5.15% |
| black girl magic | 10.07% | blue lives matter | 4.92% |
| cases | 10.07% | essential not expendable | 4.92% |
| Michigan protest | 9.84% | gentiles | 4.92% |
| communities | 9.17% | Hebrew Israelites (Black group) | 4.92% |
| white privilege | 8.28% | essential workers | 4.70% |
| how to build resilience | 7.83% | hypocrites | 4.70% |
| black girls rock | 7.61% | masks | 4.70% |
| artvsartist | 7.38% | china lied | 4.47% |
| stay home save lives | 7.38% | minimum wage | 4.47% |
| hand wara | 7.16% | real relief now | 4.47% |
| nurses | 7.16% | star wars | 4.47% |
| caucus | 6.94% | Wisconsin | 4.47% |
| families | 6.71% | women matter | 4.47% |
| stocks | 6.49% | businesses | 4.25% |
| humors | 6.26% | disparities | 4.25% |
| schooling | 6.26% | end times | 4.25% |
Fifteen topics (four themes) were identified in 1,763 publicly available COVID-19 related Tweets within African American Twitter communities (Newman modularity 0.74). Four main themes include: 1) Symptom and Transmission Patterns- cure, case, patient, death; 2) Treatments, Prophylaxis and Cures- data, CDC, mortality, rate; 3) Effectiveness/Impact of Interventions by Health Authorities and Other Institutions– day of lockdown, quarantine, covid safe, crisis, and impact; 4) Fear –protect, help, new, African American community, black, isolation, quarantinelife, and school.
4. Discussion
Social media provides a window into a community’s lived experience of the pandemic. We detected fifteen topics in publicly available Tweets by African Americans and further categorized those topics into four main themes including symptoms/transmission patterns, treatment, implementation of interventions, and fear. Importantly, positive sentiments, cohesively encouraging online discussions (e.g., Black strong 27.1%, growing up black 22.8%, support black business 17.0%, how to build resilience 7.8%), and COVID-19 prevention behaviors (e.g., masks 4.7%, encouraging social distancing 9.4%) were uniquely observed in Tweet corpus authored by African Americans.
Although repeated exposure to news and social media can be upsetting during the pandemic, the African American Twitter community has largely demonstrated healthy online coping behaviors such as giving accolades to frontline workers (e.g., warriors 16.1%, doctors 12.4%, nurses, 7.2%); keeping up their social connections as a strong and cohesive group (e.g., African American excellence 32.7%, black strong 27.1%, growing up black 22.8%, support black business 17.0%); and practicing resilience (e.g., how to build resilience 7.8%) during the pandemic.
Understanding the spread of the virus is critical because of its deadly nature. Yet, several hundred new studies arise daily [4]. A study comparing COVID-19 knowledge by misinformation source found that social media respondents were unconvinced handwashing or social distancing were valuable. In fact, tremendous distrust from the CDC emerged as social network respondents believed the COVID-19 threat was a politically motivated hoax from the CDC, the virus is man made in China and Vitamin C is a secret cure [5,6]. Inaccurate information spread on social networks generally includes the impact from COVID-19 is not real including the death toll is a lie.
Fear is evidence by people not seeking care when COVID-19 symptoms exist. Follower with inaccurate information are often confused leading to adverse effects. The US media system supports disinformation, racism and hate as it drives up the ratings and advertising revenue, making dispelling such myths even more difficult [7,8].
The generalizability of this study is limited by using the control, which did not exclude Tweets written by African American individuals. A few recent classifiers have been developed to detect demographic information in Twitter; however, they were either unavailable or in a form that is not easily applicable. As a result, some important topics uniquely discussed by African American Twitter communities may have been excluded.
5. Conclusion
Our topic modeling detected information, rumors, positive sentiments, cohesively encouraging online discussions (e.g., Black strong 27.1%, growing up Blacks 22.8%, support Black business 17.0%, how to build resilience 7.8%), and COVID −19 prevention behaviors (e.g., masks 4.7%, encouraging social distancing 9.4%) as reflections of the lived experience of African Americans during the pandemic. This new knowledge adds insights regarding future virtual based intervention and social media based health disparity research for COVID-19.
Figure 1.
Semantic visualization of topic modeling on COVID-19 Tweets authored by African Americans
Acknowledgments:
U.S. federal grant R01AG060929.
References
- [1].Yancy CW, COVID-19 and African Americans, JAMA (2020). [DOI] [PubMed] [Google Scholar]
- [2].Blei DM, Technical perspective: Expressive probabilistic models and scalable method of moments, Communications of the ACM 61 (2018), 84–84. [Google Scholar]
- [3].Freelon D, Lopez L, et al. , How Black Twitter and other social media communities interact with mainstream news, (2018). [Google Scholar]
- [4].Boyle A, Studies of coronavirus evolution stir up a controversy for scientists on social media, geekwire.com (2020). [Google Scholar]
- [5].Pennycook G, McPhetres J, et al. , Fighting COVID-19 misinformation on social media: Experimental evidence for a scalable accuracy nudge intervention, (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
- [6].Singh L, Bansal S, et al. , A first look at COVID-19 information and misinformation sharing on Twitter, arXiv preprint arXiv:2003.13907 (2020). [Google Scholar]
- [7].Ferrara E, # COVID-19 on Twitter: Bots, Conspiracies, and Social Media Activism, arXiv preprint arXiv:2004.09531 (2020). [Google Scholar]
- [8].Pakpour A,Griffiths M, The fear of CoVId-19 and its role in preventive behaviors, Journal of Concurrent Disorders (2020). [Google Scholar]