This investigation identified health influencers
Over the past years, MEAG has been organizing a small conference under the “Exploring Media Ecosystems” title with the goal of showcasing some of the best research on media ecosystems, with a focus on digital but casting a wide net in terms of topic, methodologies and geographies. We chose that specific title because it encompasses well what we are trying to do: to travel through an unfamiliar territory with the goal of learning about it, and to examine, evaluate, discuss, or inquire into something in detail.
We are starting here another exploration, a trip through unfamiliar (to us, perhaps to you) territory with the goal of inquiring into it in detail and learning about it. The territory is health communication in Africa’s media ecosystem. We have chosen a point of departure, but the path and the arrival are uncertain (we are open to suggestions and collaborations).
We started in Kenya, and set as our first goal to come up with a list of people and organizations that can be considered ‘influential’ in media coverage of health issues by Kenyan news sources. We then classified those prominent people and organizations, chose the ones that are Kenyan and that are specifically related to health, and identified their Twitter profiles. We then run an analysis of the content that they share through those profiles.
This first step in our exploration allowed us to test a specific methodology for identifying influential voices, one that we hope to replicate in other contexts and compare with alternative approaches. It also allowed to understand the range of those influential voices in the news discourse about health in Kenya. And it gave us a first view on how Twitter is used and on whether influential voices are spreading health related misinformation through it.
Below you can find a detailed account of the first stage of this exploration.
We began by identifying individuals who are influential in the topic of health in Kenya. There are a number of ways to scope and define influencers in public discourse, particularly when social media is taken into account. For this study, we undertook a novel approach that centered on the individuals and organizations mentioned most frequently in news media stories on the topic of health.
Using Media Cloud, our open source database of global news media, we pulled stories from our collection of Kenyan media sources publishing news at a national level. There are 67 unique media sources in this collection.
We scoped first to stories published between January 1, 2019 and November 15, 2020, a total of 395,365 stories. We then narrowed to stories that were detected by a trained theme classifier as pertaining to the topic of health. This theme classifier was developed using transfer learning - starting with the Google News word2vec models and then adapting them to produce tags based on the New York Times annotated corpus. We score each story against the most common 600 descriptors from the NYT corpus. Any descriptors that score above 0.2 probability are listed as theme(s) for the story. For more technical details about the development of this classifier, please see Yasmin Rubniovitz's MS thesis (section 4.3). Narrowing to health-themed stories resulted in 63,576 stories (16% of the original corpus).
We then used Media Cloud’s CLIFF-CLAVIN module to extract entities mentioned in these stories. CLIFF-CLAVEN ingests entities through Stanford's Named Entity Recognizer, one of the main entity recognition models in use in computer sciences today, and then adds some post processing to disambiguate place names only. CLIFF-CLAVIN produced a list of 931 individuals and 1000 organizations (the upper limit of reporting), with the story mentions ranging from 17 (0.03% of stories) to 5645 (8.8%). We set a threshold of 50 story mentions, excluding entities that were mentioned in 49 stories or fewer. After cleaning and deduplication of the entity names, we arrived at a list of 418 entities.
Researchers then used desk research to identify the position/description of each entity. This involved both looking up the name of the individual or organization online, as well as revisiting the news stories that included the entity and reading related stories for context. After each entity’s description was noted, we then manually coded entities into the following categories:
For this study, we focused on individuals in the category of Kenyan health. Of the 57 entities in this category, two were excluded for further analysis on the basis that they were in the news on health topics only for being the subject of an investigation into fraud at the Kenya Medical Supplies Authority (KEMSA). For the remaining 55 entities, we set out to collect the Tweets made by these individuals, and analyze the digital media they shared on the platform.
We first undertook desk research to identify the Twitter handles of the Kenyan health entities. 14 of the 55 entities were not found to have a Twitter handle. For an additional four of the entities, there were too many Twitter handles with the given name and no way to identify which of the handles, if any, authentically belonged to the entity in question. The remaining 37 entities had identifiable Twitter handles, listed next to the individual/organization and their description below. We also note whether or not the entity’s Twitter handle is verified by the Twitter platform. Interestingly, only 11 (30%) of these influential individuals were verified. Finally, six accounts did not create Tweets within the last two years, noted with an asterisk in the table. These accounts were therefore not included in Tweet analysis, resulting in a final corpus of 31 active accounts.
We then used the Twitter API v2 to collect the Tweets made by these accounts during the two-year period of January 01, 2019 through December 31, 2020. This resulted in 41,478 Tweets. Filtering for only Tweets that contained hyperlinks to non-Twitter media URLs, we arrived at 1,766 Tweets. As several Tweets contained more than one link, the total number of links came to 1,787. We then analyzed these links, described in the findings below.
There were 339 unique domains shared within the Tweets from the health influencers. After eliminating the URLs that were shared as an entity linking to its own website, the top most frequently shared domains are shown in the bar graph below.
YouTube was by far the most frequently shared site, with more than twice the number of shares of the next most popular domain, Kenyan news site Nation. Facebook links were the third most shared. Social networking sites Periscope.TV and Instagram also made it to the top 20.
News sites make up the largest category in the top shared domains (50%). The most shared news domains are Kenyan: Nation, The Standard, The Star, and Business Daily Africa. Lower down the list are international news outlets New York Times, the Guardian, and the BBC, along with journals The Lancet, Science, and Nature.
The regional health authority, Africa CDC, is in the top 20, at number 10, as is the World Health Organization and the African Union. National health authority the Ministry of Health does not make the top 20.
There were 1,394 unique URLs shared within the Tweets from the health influencers (with a total number of posts of 1,787). After removing self-referential content, the top 20 most frequently shared URLs are shown in below.
The homepage of the Kenyan news outlet Nation was by far the most shared URL, but this appears to be due to Twitter sharing links from Nation resolving to the homepage when visited. Kenyan news outlet The Standard had two specific articles in the most shared URLs. Two scientific articles about COVID-19 made the most shared list; other health condition-content among the top include a mental health report from the Ministry of Health, a heart attack informational video from Capital FM, and a story about breast cancer from The Standard. Interestingly, several of the most shared URLs have to do with the industry of medical practice, such as articles on telemedicine and articles highlighting challenges of health care providers (for example, the two articles from the Health Workers for All Coalition).
YouTube was by far the most shared domain in Tweets by the health influencers. Of the 234 URLs shared, only 23 were shared more than once. The YouTube video shared most frequently was a segment from Capital FM explaining the signs of heart attacks (5 shares). Three videos were shared second most frequently, with three shares each: a music video, a Citizen TV segment about the first Kenyan patient to survive COVID-19, and a video from a Kenyan Human Rights Commission event.
YouTube videos ranged from news segments, to content produced by health organizations themselves, to individuals talking and sharing their views about the news of the day. None of the YouTube videos appeared to be dis- or mis-information content. By and large the content was information and teaching content, with 30% being news clips. 73% of the videos focused on a health issue; of these, the most significant (28%) was coronavirus-related content.
Only three of the 91 Facebook URLs shared were posted more than one time; these were links from the Kenya Medical Association and the Kenya Medical Training College sharing their own content. While more political content was shared from the Facebook domain than others, spot checks of the content did not surface any clear health misinformation.
Of the 37 Periscope.TV URLs shared, only five were shared more than once (all were shared twice). The Periscope content was primarily videos created and shared by the health organizations Kenyan Medical Practitioners Pharmacists & Dentists Union (KMPDU), Kenyan Medical Association, and the Kenyan Human Rights Commission (KHRC). One of the videos shared twice was from NTV news.
Nearly all of the links to Instagram (10 out of 12) were shared by the National Cereals and Produce Board, linking to their own Instagram account. The Instagram login page was posted by four authors, and the Kenyan Medical Practitioners Pharmacists & Dentists Union (KMPDU) shared one link to a post on their Instagram account.
Through this work we have sussed out several interesting facets of the health media ecosystem in Kenya. These include:
It should be noted, encouragingly, that the Kenyan health influencers seem to be using Twitter to post relevant and non-harmful content. Through our exploration we did not find any instances of these influencers sharing health misinformation.
We plan to continue this exploratory research into identifying health influencers in African nations, and seeing what we can learn about media shared in this ecosystem. Let us know what you think we should dig into next!