The role of associations in content shareability.
In light of the coronavirus pandemic, content plays a crucial role in informing and protecting the public. Coronavirus headlines and images elicit associations that are shaping which content is being shared and how people understand the crisis. Unfortunately, this applies to real and fake news [0].
In this post, you will learn about the associations that are connected to coronavirus headlines and images, ultimately making content more shareable and viral.
Executive Summary
- Headlines that triggered associations with “generous”, “encouragement”, “safety” and “accessibility” were shared more.
For example, Headlines that talk about people donating money for fighting the coronavirus, experts giving advice on how to be safe, answers to questions regarding the accessibility of essential goods and services. The message doesn’t need to be positive, instead, content that addresses these topics, in general, is shared more.
- Images were shared more when they featured symbols of authority.
Pictures in which people are wearing format clothes, white lab coats or other symbols of status and expertise were shared more. We believe that in times of uncertainty, people look for reassurance and clarity from people who they perceive to be in charge. Indicators of authority are an efficient mental shortcut people use to accept information from that source.
- Images were shared less when they showed people wearing face masks.
We believe that face mask wearers are seen as being sick and thus pose a risk that needs to be avoided. Face masks are also associated with “fear” in the English language.
Check out the rest of the analysis to understand why we believe that, things that appear to threaten or support “the tribe” are shared more.
Methods for Analyzing Coronavirus Headlines and Images
(Left) Images of coronavirus articles. (Right) The most frequently used words used in the Coronavirus headlines. Larger words appeared more often.
We looked at 11.333 headlines and images from western news websites, published between the fifth of December 2019 and 11th of March 2020, using the Aylien News API.
To make shareability between news sources comparable, we calculated “share velocity”. To illustrate, a news article with a share velocity of 5, is shared 5 times faster than a typical article on that site.
The western news sources for coronavirus headlines and images, with the respective count and source name.
We then used a variety of methods to calculate the relationship between associations and the number of shares the content received.
Viral Content Examples of Coronavirus Headlines
Representation of which coronavirus headlines were most viral across the various successful associations. A score of 255 means that that post was 255 times more shared than a typical article from that particular website.
Text Associations in Coronavirus Headlines
The Associations of Shareable Content
Previous studies found negative connotations to be predominant in public health messages [3]. Surprisingly, we found that Coronavirus headlines evoking positive associations of generous, encouraging, accessible were shared more often.
This means that more viral headlines featured one or several of these associations.
Why Are These Associations Connected With More Shares?
While there are multiple drivers of a topic’s virality, we find that the strongest correlates are themselves positive. But that alone may be misleading.
Many of the headlines evoke the potential for our safety being under threat or talk about medical equipment not being accessible. So while the topics themselves have predominantly positive associations, the reason why they are connected to more shares may very well be because the topic is of high relevance overall.
To test this assumption, we analyzed whether Encouraging words are related to words like hope and normal or resolution. As you can see in the image below, those words are indeed highly related.
So why are the viral headlines driven by the topics of Generous, Safety, Accessible, and Encouragement? Possibly because these topics capture the need for normality and are linked to the potential to feel hope. They are also connected to a sense of agency and progress that might be inspiring to people in times of uncertainty.
Exploring what drives the viral text associations.
Image Associations:
Representation of images we analyzed to identify associations of virality in images of coronavirus articles.
Images of Authority Are Shared More: Visual heatmap analysis
The heat maps show which parts of the picture contribute the most to the article’s shareability. Thanks to NVIDIA and the Telekom Techboost for supporting the analysis with their Cloud Infrastructure.
The parts of the images that made the neural network trigger success, showed people in formal outfits. This is in line with a well known social influence principle in psychology, appeal from authority.
Appeal from authority means that people search for indications of expertise, such as titles and uniforms, lab coats and suits. Once such indicators are detected, people are more likely to take the depicted person’s opinion seriously.
People are also more likely to comply with requests when given by a person wearing formal clothing [6].
Thus, once authority indicators are recognized, people appear to use heuristics like “this person wears professional clothing, they must be a professional” and then adjust their personal beliefs to bring them closer to the expert opinion [7].
In our analysis, content that depicted authority figures were shared more often. Therefore, when the state or journalists want to spread health information more effectively, we recommend using authority indicators. To name a few: The name of a respected institute or university, people wearing suit and tie, physicians wearing lab coats, certificates, and logos of well-known brands.
Trump’s press conference on 13th March 2020 featured well-dressed CEOs of large brands. A quick glance at the picture below shows several cues of authority that shape people’s understanding of how trustworthy the information they are getting might be.
Representation of images we analyzed to identify associations of virality in images of coronavirus articles.
Image Credit: Chip Somodevilla/Getty Images, taken from [link]
Associations of Objects on Images
We also looked at the objects identified on the more shareable images. The associations of objects on the image, with a strong relationship with more shares, were:
In line with the heatmap analysis, we find associations relating to authority (Confidence, Fearlessness, Decisive etc.) to be correlated with more success.
Image with strong “Decisive” associations (Objects: White House, suits, red tie, Donald Trump, flag.)
Associations Don’t Need to Be Positive
Note that we are talking about associations here. A headline like “all confidence is lost, the best scientists have failed us, and we honestly don’t know what to do now” would likely be shared as well, as it is associated with confidence, fear, and honesty.
The factors behind sharing coronavirus content are related to the topics that are of the highest relevance right now. It is the relationship between association and shareability that content writers should mind, to understand what content is more likely to be shared.
Now we turn our attention to factors related to below-average shareability.
Images of Asians with Face Masks are Shared Less
What group of people do we associate with the coronavirus? What do we think when we see someone wearing a head mask?
On February 11th, 2020, the WHO announced that the 2019 novel coronavirus would thereon be referred to as COVID-19. The current Director-General, Tedros Adhanom, explained how the WHO followed the guidelines to ensure the name does not refer to a geographic location, an animal, an individual or a group of people.
Heat maps showing parts of each picture that triggered the failure prediction. The pictures above were found to have below-average number shares.
In spite of this effort, the reports on racially motivated attacks and acts of prejudice against people of Asian descent have spiked in the Western world.
In Italy, this ethnic group has experienced targeted aggression, insults, dirty looks, and jokes [10].
In Germany, people of Asian descent have experienced racism and refusal of service [11].
In the US, there have been reports of physical and verbal harassment, and Chinese restaurants have even experienced a sharp decline in sales [12].
But, how can this negative social attention be identified and pinpointed in the digital world?
Our heatmap analysis shows that the image area predictive of below-average shareability depicts people of East-Asian descent wearing face masks.
This observation might be a consequence of Western people being less familiar with Asian faces, making it easier to see them as “other” and as members of an out-group. This phenomenon may be supported by westerners’ practice of “whitewashing”. The practice of casting white actors in the roles of people from Asia [13].
Such practices increase and reinforce the familiarity of white faces. Since exposure is linked to liking and preference a phenomenon known as the mere-exposure-effect [14]), provides an explanation of why such images may be less liked and shared.
Another notion to explore here are the associations that people hold for face masks. The technology we have developed for our software is able to predict what associations people hold for any word or topic. Looking at the results, we found a strong association between face masks and fear:
Heat maps showing parts of each picture that triggered the failure prediction. The pictures above were found to have below-average number shares.
The association with Face Mask and Fear is considerably higher than it is for other head/facial garments like Turban and Bandana.
This could perhaps be understood through the assumption that the face mask wearer might be sick and thus pose danger. While the reason behind wearing a face mask might indeed be illness, many healthy people chose to wear it in order to protect themselves from infections. Such nuances, however, are often not taken into account when anxiety levels are high.
In fact, when people feel threatened, they instinctively turn toward their in-group (those that look, act, and think like them) and away from the out-group (towards whom they are more likely to display signs of aggression) [15].
It is then not surprising that we witnessed such horrible acts of aggression during times of great fear. With regards to COVID-19 and hate-crimes against people of Asian descent [12], we’d like to conclude this section with the wise words of the WHO Director-General Tedros Adhanon:
“Stigma is the most dangerous enemy. For me, it’s more than the virus itself” [16].
Conclusion
We analyzed thousands of Coronavirus headlines and images about the coronavirus outbreak. Using machine learning, word embeddings and neural network activation mapping we determined the factors connected to higher shareability of coronavirus related content.
Taking a bird’s eye view, we found:
Authority is key in times of uncertainty: Shareability of image content appears to be linked to indicators of authority and status. Pictures of people in formal attire, suits, and lab coats, were shared more often than those with people in casual clothing.
We suggest that in times of uncertainty, people look for reassurance and clarity from people who they perceive to be in charge. Indicators of authority are an efficient mental shortcut people use to accept information from that source.
Hope for the tribe being strong: News about people who are supporting the community, e.g. by donating money or pledging to support initiatives, are like beacons of hope in dark times. Since humans are a species that survives best when acting as a group, seeing people who put the needs of the community above their own is a powerful social signal.
We believe that this communicates that there are still resources to be shared, that cooperation is still possible and that there is hope that things will be better.
Hope can be a powerful driver. Good leaders communicate difficult topics with honesty and ensure that people still feel like they belong to a united community. A community that can come together in hours of need to stand as one.
With dark topics on everyone’s mind, people share information that not only informs about the situation but also helps answer the question of how to feel right now.
There is much potential for anxiety to build and people to see more problems than solutions. Previously we presented further evidence that difficult topics, like famine, refugees and discrimination are more engaging, when hope is also communicated [UNICEF Facebook Analysis].
In times of trouble, people are likely looking for want certainty. We believe that news articles that carry authority, depict the better side of humanity, and encourage a notion of hope and unity. Reaching more people is crucial.
Convincing people that avoiding social contacts or reducing it to the minimum to slow the spreading of the disease is crucial for our society right now. However, there are still many out there not convinced this is necessary or believing they are invincible.
Remember, your writing can make a difference in how people feel, think and act. Use words and images wisely.
Fake News Warning
The same principles that lead to more shares for real news can also be used to spread fake news. For example, a WhatsApp message that cites the University of Vienna (appeal to authority) in which the use of Ibuprophene is discouraged, was shared thousands of times even though it was completely fake.
The University of Vienna, fighting against Fake News
This illustrated that it is often better to trust original sources over shared content on other media.
If you want to learn more about the facts, consider this comprehensive summary of coronavirus information.
Acknowledgments
This analysis was the joint effort of the Data Science and content team at NEURO FLASH.
Special thanks to Pio Calderon for the data analysis and interpretation. Ana Zdravic and Daniel Anadria for the research and writing most of the text. Artus Krohn-Grimberghe from Lytiq for supporting the image analysis. The Telekom Techboost Program and NVIDIA for supporting us with their Cloud infrastructure (CPU and GPU Servers).
References
[0] https://www.politico.com/news/2020/03/16/coronavirus-fake-news-pandemic-133447
[1] Kott, A., & Limaye, R. J. (2016). Delivering risk information in a dynamic information environment: Framing and authoritative voice in Centers for Disease Control (CDC) and primetime broadcast news media communications during the 2014 Ebola outbreak. Social Science & Medicine, 169, 42-49.
[2] Rousseau, C., Moreau, N., Dumas, M. P., Bost, I., Lefebvre, S., & Atlani-Duault, L. (2015). Public media communications about H1N1, risk perceptions and immunization behaviours: A Quebec–France comparison. Public Understanding of Science, 24(2), 225-240.
[3] Lee-Won, R. J., Na, K., & Coduto, K. D. (2017). The effects of social media virality metrics, message framing, and perceived susceptibility on cancer screening intention: The mediating role of fear. Telematics and Informatics, 34(8), 1387-1397.
[4] Bassett, R. E., Staton‐Spicer, A. Q., & Whitehead, J. L. (1979). Effects of source attire on judgments of credibility.
[5] Rehman, S. U., Nietert, P. J., Cope, D. W., & Kilpatrick, A. O. (2005). What to wear today? Effect of doctor’s attire on the trust and confidence of patients. The American journal of medicine, 118(11), 1279-1286.
[6] Miller, F. G., & Rowold, K. L. (1980). Attire, sex-roles, and responses to requests for directions. Psychological Reports, 47(2), 661-662.
[7] Cialdini, R. B. (2009). Influence: Science and practice (Vol. 4). Boston, MA: Pearson education.
[8] The World Health Organization. (2020, February 11). WHO Director-General’s remarks at the media briefing on 2019-nCoV on 11 February 2020. Retrieved March 14, 2020, from https://www.who.int/dg/speeches/detail/who-director-general-s-remarks-at-the-media-briefing-on-2019-ncov-on-11-february-2020
[9] Serhan, Y., & McLaughlin, T. (2020, February 13). The Other Problematic Outbreak – As the coronavirus spreads across the globe, so too does racism. Retrieved March 14, 2020, from https://www.theatlantic.com/international/archive/2020/03/coronavirus-covid19-xenophobia-racism/607816/
[10] Giuffrida, F. (2020, February 26). L’incubo di essere cinesi in Italia con il coronavirus: «Un ragazzo preso a bottigliate in Veneto». Retrieved March 14, 2020, from https://www.open.online/2020/02/26/lincubo-di-essere-cinesi-in-italia-con-il-coronavirus-un-ragazzo-preso-a-bottigliate-in-veneto/
[11] Shioida, T. (2020, March 1). „Die Plätze neben mir in der U-Bahn sind immer leer“: Eine Japanerin erzählt, wie das Coronavirus das Leben von Asiaten in Deutschland verändert hat. Retrieved March 14, 2020, from https://www.businessinsider.de/leben/die-plaetze-neben-mir-in-der-u-bahn-sind-immer-leer-eine-japanerin-erzaehlt-wie-das-coronavirus-das-leben-von-asiaten-in-deutschland-veraendert-hat/
[12] Yan, H., & Chen, N. (2020, February 21). What’s spreading faster than coronavirus in the US? Racist assaults and ignorant attacks against Asians. Retrieved March 14, 2020, from https://edition.cnn.com/2020/02/20/us/coronavirus-racist-attacks-against-asian-americans/index.html
[13] Lee, T. G., & Gandhi, L. (2017, December 21). Hollywood has whitewashed Asian stories for decades. This year, they couldn’t ignore the backlash. Retrieved March 14, 2020, from https://www.nbcnews.com/news/asian-america/hollywood-has-whitewashed-asian-stories-decades-year-they-couldn-t-n830241
[14] Zajonc, R. B. (1968). Attitudinal effects of mere exposure. Journal of Personality and Social Psychology, 9(2, Pt.2), 1–27. https://doi.org/10.1037/h0025848
[15] Wanis, P. (2017, August 22). The Psychology of Hate. Retrieved March 14, 2020, from https://www.patrickwanis.com/the-psychology-of-hate/
[16] Jarasevic, T., Ghebreyesus, T., Ryan, M., Van Kerkhove, M., & Questioners. (2020, March 2). WHO press conference on COVID-19. Retrieved March 14, 2020, from https://www.who.int/docs/default-source/coronaviruse/transcripts/who-audio-emergencies-coronavirus-press-conference-final-02mar2020.pdf?sfvrsn=cf76053d_2
