The PRISMA framework for systematic reviews (2009) recommends identifying material for review from medical databases and “other sources”. The traditional approach is to look at “grey literature” – government reports, lay press, articles in publications other than peer-reviewed scientific journals. Increasingly such reports and publications are available online, either direct through a publication’s website or found through a search engine. Unfortunately search engine outputs can be unpredictable (e.g. in terms of ranking and completeness), even when searching specifically for “news”. This may be because of the way that new stories are tagged by the news source, or by idiosyncrasies in the search engine algorithms which are black-box/ proprietorial. 
News outlets and citizen journalists also promote their output through social media. Twitter is the most immediately accessible source of social media information: it is a largely open platform, with only a minority of tweeters having private accounts (~ 12%); it serves up content in a relatively predictable fashion; and data can be accessed easily through the Twitter API, either directly or through third-party applications. One such application – NodeXL – provides access to up to 18,000 tweets and retweets per download, providing a detailed web-based report of tweeters and their connections, as well as access to raw data. Such information can be curated, e.g. by selecting content based on the number of retweets (Twitter terminology for dissemination of content, e.g. because a tweeter has found the content interesting or controversial, seeking feedback). Many tweeters will post content without receiving retweets, while other content will become viral, and will dominate Twitter displays of such content (e.g. in the Twitter “top” tab for searches, or “while you were away” lists if Twitter has identified that you are already interested in a topic due to previous activity). 
There are different ways to prioritise news stories on Twitter using raw tweet data. One is to look at the number of tweets (i.e. original post) that reference a URL to the news story; an alternative is to look at the number of times these tweets have been shared through retweeting. The first approach documents the intention to share original information, while the latter provides a proxy measure of the reception of that information. NodeXL ranks content using both information. One complication, however, is that tweeters will frequently shorten weblinks (URLs) to news stories that they quote, e.g. through bit.ly and tinyurl, and indeed Twitter itself shortens URLs. Fortunately, NodeXL expands URLs during data extraction, providing automated collation of references to a news story(*). The analysis presented here is based on a series of NodeXL extracts looking at searches for articles about COVID-19 and meat processing plants. The searches were based on broad search strategies that were viewed initially in a rapid web-based tool, and refined. Twitter search terms have to be specific – Twitter does not use approximate string matching (“fuzzy search”). NodeXL extracts on this scale can take hours to complete, and have to be run from a powerful PC. This can be challenging to schedule in addition to a busy work schedule. Extracts were left to run overnight, and while at work. As additional terms were added to searches (e.g. “hog” or “hogs” instead of “pig” or “pigs”), other terms were removed using the equivalent of Boolean operator in Twitter (adding the minus sign in front of a search term). Viewing the results of the searches, a future search could have excluded terms such as “Elon Musk” and “guinea pig”, as they were not related to our research question, but appeared in terms related to factories (Musk) and COVID-19 testing (guinea pigs). 
NodeXL is primarily a social network analysis tool, looking at the connections between tweeters. Views on meat and meat processing (e.g. worker rights, working conditions, unions), and regulations around public movement during the COVID-19 pandemic are likely to be polarised at a societal level, potentially along party political lines. Similar patterns are seen with other public and environmental health measures, e.g. vaccination, tobacco control, climate change. NodeXL maps different communities based on tweet content, interactions in the extracted tweets and other data, and displays information about top content (up to 10 URLs) for the most influential clusters (up to 10 clusters). For topical issues this gives a broader range of information than focusing on the top content alone. Such content can be viewed by clicking on any of the NodeXL searches produced for this analysis (e.g. this search for chickens). With the “community clusters” and fragmented tweeting that we observed in the NodeXL maps that we had already run, we quickly confirmed that we were identifying a useful spread of information for generating themes for further exploration. 
Given the limits on number of tweets and retweets (up to 18,000 per extract) and time period (up to 7-10 days into past), a series of searches were run between 14 and 18 May 2020 to capture as complete a record as possible for the period 6 to 14 May 2020 inclusive. Individual NodeXL reports were read, viewing the website of promising material, typically on the basis of the wording of the URL – e.g. in report above one link, shared 89 times (a figure recorded in square brackets in the NodeXL report), has the URL https://www.wired.com/story/why-meatpacking-plants-have-become-covid-19-hot-spots/, making the content clear. Other popular material without a clear title was also viewed (typically at least 5 shares). 
The topics from relevant articles were recorded and themed, grouping articles. This information was used, in turn, to refine subsequent searches. Hashtags were also inspected in the NodeXL report to see if the topic was being tagged by terms that could help adapt searches further, but no unifying terms were identified.

Items