Why did it go viral? An informatics-based case study of exaggerated language in news and social media
Plain Language Summary
News stories about science sometimes use exaggerated words that were not used by the scientists themselves, words like ‘astonishing’ or ‘miracle’, for example. Words like these might lead to a science story 'going viral' across social media and other new sites but could also lead to the public being misled and might foster distrust in news stories.
In this study, we looked at the viral spread of a science news story that was presented at the American Society of Clinical Oncology Annual Meeting in June 2022 and published at the same time in a prestigious scientific journal (the New England Journal of Medicine). It was about a new treatment (called dostarlimab) for a specific kind of rectal cancer. Despite involving only 12 patients, the study led to a surprising amount of news and social media interest.
We used social media and news monitoring software to find posts and articles about the study, and we used artificial intelligence tools to get insights into what the first articles were and whether they used exaggerated language.
We found a few early articles from major news outlets that included exaggerated language (for example, ‘revolutionary’, ‘unprecedented’, ‘scientific miracle’). However, these were not picked up on social media until a few days later, when they were cited in several highly influential tweets from politically affiliated accounts. These early news articles were also picked up by other news articles that used similar exaggerated language, which may have triggered the high levels of public interest and the viral spread.
We think that making plain language summaries of scientific publications available to the public might help to lower the risk of exaggerated reporting in the future.
In this study, we looked at the viral spread of a science news story that was presented at the American Society of Clinical Oncology Annual Meeting in June 2022 and published at the same time in a prestigious scientific journal (the New England Journal of Medicine). It was about a new treatment (called dostarlimab) for a specific kind of rectal cancer. Despite involving only 12 patients, the study led to a surprising amount of news and social media interest.
We used social media and news monitoring software to find posts and articles about the study, and we used artificial intelligence tools to get insights into what the first articles were and whether they used exaggerated language.
We found a few early articles from major news outlets that included exaggerated language (for example, ‘revolutionary’, ‘unprecedented’, ‘scientific miracle’). However, these were not picked up on social media until a few days later, when they were cited in several highly influential tweets from politically affiliated accounts. These early news articles were also picked up by other news articles that used similar exaggerated language, which may have triggered the high levels of public interest and the viral spread.
We think that making plain language summaries of scientific publications available to the public might help to lower the risk of exaggerated reporting in the future.
Introduction
- News stories and social media coverage about medical innovations can be exaggerated1,2 leading to ‘viral’ stories that may erode public trust in science and medicine.
- We selected a case study to examine this phenomenon: a study of a novel rectal cancer treatment in 12 patients presented at the 2022 American Society of Clinical Oncology (ASCO 2022) Annual Meeting, and simultaneously published in NEJM.3
- This publication generated extremely high news and social media interest, and so provides a rich source of media content.
Objective
- We sought to understand the triggers and independent channels of the dissemination of this study and how the viral language used by these independent channels differed from the scientific language used by the investigators/authors in the originals.
A presentation at ASCO 2022 generated very high interest in rectal cancer
How this was discovered
- Google trends data ‘rectal cancer’ in the period 30 November 2019–30 November 2022.
Influential tweets came from politically affiliated accounts several days after initial publication
Interactive retweet network plot can be further explored by:
- hovering over nodes to display Twitter handles
- hovering over edges to view dates and timings of retweets
- clicking on nodes to highlight connections (retweets) between Twitter handles.
How this was discovered
- Twitter search for ‘dostarlimab’ and ‘rectal cancer’ (date range 5 June 2022–7 July 2022). Network cluster analysis based on retweets.
Media sources cited by key tweets used exaggerated language
The New York Times
NPR
Clustering of news/web articles found large numbers that shared the language of the media sources, and also ones that were more similar to the language of the NEJM abstract and editorial
Major news article clusters
Click on the labels in the figure legend to compare article counts across different article clusters.
Cluster B1 contains articles similar to the NY Times article while clusters B2 and C2 contain articles similar to the NEJM abstract and NPR article, respectively.
Cluster B1 contains articles similar to the NY Times article while clusters B2 and C2 contain articles similar to the NEJM abstract and NPR article, respectively.
How this was discovered
- Feedly search for news/web articles on ‘dostarlimab’ and ‘rectal cancer’ (date range 5 June 2022–5 July 2022).
- Key articles that were not found in the search were manually downloaded.
- Topic clustering was performed using BERTopic.
Comparison of word use between clusters of articles related to the media sources showed that they shared exaggerated language that is less common in more scientific sources
Use the dropdown menu to compare word use between different article clusters.
Cluster B1 contains articles similar to the NY Times article while clusters B2 and C2 contain articles similar to the NEJM abstract and NPR article, respectively.
Cluster B1 contains articles similar to the NY Times article while clusters B2 and C2 contain articles similar to the NEJM abstract and NPR article, respectively.
How this was discovered
- Text cleaning removed non-English articles and standard stop-words and performed lemmatization.
- Network analysis of influential terms identified several articles with large amounts of irrelevant text – these articles were excluded from further analysis.
- Part-of-speech (pos) tagging was used to identify adjectives.
Conclusions
- We found that news articles from major publishers used exaggerated language that was picked up by other media and by influential non-medical social media accounts.
- Exaggerated language used in some news articles, which substantially overemphasized the narrative of this trial, likely played a key role in the viral spread of information among the general public in news and social media.
- Plain language summaries can enable access to accurate, fair and balanced interpretation of medical research,4 which may counter the perils of exaggerating or extrapolated reporting.
- This AI approach combining natural language models with text analytics allows analysis of large volumes of text to gain insights into the spread of messages.
References
- Sumner P et al. BMJ 2014;349:g7015.
- Suarez-Lledo V, Alvarez-Galvez J. J Med Internet Res 2021;23:e17187.
- Cercek A et al. N Engl J Med 2022;386:2363–76.
- Pushparajah DS et al. Ther Innov Regul Sci 2018;52:474–81.
Disclosures
HR, TR, AL: Employees of Oxford PharmaGenesis. LK: Nothing to disclose. AP: Employee of Novartis with no direct financial interest related to the assets or topics discussed in this poster.
Poster number: 23
Presented at the European Meeting of ISMPP | 24–25 January 2023
Presented at the European Meeting of ISMPP | 24–25 January 2023