George Mason University
George Mason University Mason
George Mason University

Exploring Social Media’s Potential to Improve Public Health

December 4, 2019   /   by Michelle Thompson

In a world where tweets literally travel faster than seismic waves, it is imperative to know when a wolf really is a wolf — and when it’s not.  Together, an epidemiologist and a computational social scientist unleash the power of Web 2.0 data.

social media iconsDrs. Andrew Crooks and Kathryn H. Jacobsen see social media through a different lens than your average Twitter or Instagram user. Crooks, a computational social scientist in Mason’s College of Science, and Jacobsen, an epidemiologist in Mason’s College of Health and Human Services, study how social media influences public health perceptions and behaviors as well as how public health information is disseminated.

The vast data sources created by Web 2.0 – through sites that promote features such as crowdsourcing and location-sharing—provide researchers like Crooks and Jacobsen insights into how social systems work and the patterns in how information spreads. More than 350 million tweets are sent per minute, and 75% of all users include some sort of geolocation information in their tweets. Crooks explains that computational social science (CSS) allows us to harness that data to explore the world around us, study the connections between people, derive knowledge about human behavior, and ultimately use that knowledge to improve public health. [The data also reveal quite a bit about Bot behavior—more on that later.]

Kathryn Jacobsen

Computational social science is crucial in a complex adaptive system where the behavior of an individual actor does not convey an understanding of the whole system’s behavior. “Yes, data can show patterns, but we need to understand the context and the process behind that data,” says Crooks, who calls upon a familiar fable to illustrate his point. “If one boy cries wolf, there probably isn’t a wolf. If 100 people cry wolf, there’s probably a wolf. If 100 people cry wolf and post pictures of the wolf at that geolocation, there’s definitely a wolf.” To make sense of social media narratives, as illustrated by the wolf example, we need three critical pieces of information: actors, concepts, and locations. 

Even with machine learning capabilities, truly understanding the “concepts” behind the data requires significant manual labor. Take the example of the extensive work to code or classify the sentiment behind social media posts on vaccinations—a controversial and politicized topic on social media. “How would you code a tweet that says ‘I don’t feel anyone should be forced’ to vaccinate.  Is that pro-vaccination, anti-vaccination or neutral?” Crooks asked in a recent Health Administration Policy Seminar. [The correct answer is anti, in case you’re wondering.][1] Providing that context is critical in understanding the nature of the discussion and tracking patterns in how the information flows across systems.

After arduous manual data classification, the research team was able to deploy machine learning analysis of more than 660,000 tweets about vaccinations and found, to their surprise, that 75% of the traffic on social media was pro-vaccination.  Their analysis showed that those who were tweeting anti-vaccination sentiment were largely talking among themselves and there was little engagement between anti- and pro-vaccination social media users. 

Andrew Crooks

Although these emerging types of data are valuable sources for researchers, it can be expensive to acquire, store, and analyze social media data. For example, Instagram data used to be free via open application programming interfaces (APIs), but Facebook, which now owns the platform, now assesses hefty fees that can be in the $100k range.  “If you are tracking the data yourself,” Crooks reports, “you may be able to avoid licensing fees, but you still have the challenge of storing huge amounts of data.” 

In 2015, before the widespread outbreak of Zika virus in the Americas, Jacobsen gave Crooks a heads up that he may want to start collecting social media data about Zika. Based on her tip, they were able to watch the narrative unfold across social channels—seeing who was spreading information, where the clusters of discussion were forming, and the frequency of words being associated with one another. They were also able to see that the U.S. Centers for Disease Control and Prevention and World Health Organization were leading the flow of scientific information about Zika. [2] That was different what they observed in their research on vaccines, for which the flow of information was primarily driven by media outlets.[3] 

In a world where tweets literally travel faster than seismic waves, it is imperative to know when a wolf really is a wolf and when it’s not—and to understand how information and misinformation spreads.  Interdisciplinary collaboration between Crooks, Jacobsen, and their colleagues will continue to play a critical role in harnessing social media - and the growing mass of data - to improve public health.

A Note on The Role of Bots in Social Media Narrative

Chatbots or Bots, computer programs designed to simulate human conversation, present an additional challenge in understanding the flow of information between virtual and physical neighborhoods. To show the magnitude of the challenge, studies report that 15,000 Twitter accounts identified as Bots used the #IVoted hashtag on election day during the 2018 elections. While automated Bot detection services do exist, there is little consensus across services when it comes to classifying which specific Tweets are generated by a Bot and which are generated by a human – making manual review important.

 

[1] Yuan, X., & Crooks, A. (2018). Examining Online Vaccination Discussion and Communities in Twitter, Proceedings of the 9th International Conference on Social Media and Society, Copenhagen, Denmark, pp.197-206.

[2] Stefanidis, A., Vraga, E., Lamprianidis, G., Radzikowski, J., Delamater, P.L., Jacobsen, K.H., Pfoser, D., Croitoru, A., & Crooks, A. (2017). Zika in Twitter: Temporal Variations of Locations, Actors, and Concepts JMIR Public Health Surveill, 3(2):e22, DOI: 10.2196/publichealth.6925

[3] Radzikowski, J., Stefanidis, A., Jacobsen, K.H., Croitoru, A., Crooks, A., & Delamater, P.L. (2016). The Measles Vaccination Narrative in Twitter: A Quantitative Analysis, JMIR Public Health Surveill 2(1):e1, DOI: 10.2196/publichealth.5059

[4] Vraga, E., Stefanidis, A., Lamprianidis, G., Croitoru, A., Crooks, A., Delamater, P., Pfoser, D., Radzikowski, J., & Jacobsen, K. (2018) Cancer and Social Media: A Comparison of Traffic about Breast Cancer, Prostate Cancer, and Other Reproductive Cancers on Twitter and Instagram, Journal of Health Communication, 23:2, 181-189, DOI: 10.1080/10810730.2017.1421730

[5] Vraga, E., Radzikowski, J., Stefanidis, A., Croitoru, A., Crooks, A., Delamater, P.L., Pfoser, D.,  Jacobsen, K.H. (2016). Social Media Engagement with Cancer Awareness Campaigns Declined During the 2016 Presidential Election, World Medical Health Policy, DOI 10.1002/wmh3.247

 

 

Schools & Programs