Thematic Networks amongst the Sharers of Problematic Information on Facebook

The final paper in this ECREA 2022 session is presented by my colleague Dan Angus, and explores the sharing of mis- and disinformation on Facebook as part of our current ARC Discovery project. Our objectives are to identify and categorise the Facebook spaces that are sharing such problematic content, and the themes that they address in their sharing. This might also identify the interconnections and overlaps between such themes and topics, and the way that such connections change over time, especially with the impact of COVID-19 and other major disruptive events.

Here are the slides for this presentation, and my liveblog of Dan’s presentation follows below:

For this research, we iteratively developed a masterlist of problematic news domains called FakeNIX, building on previously published research; this now contains some 2,300 domains. We then used Facebook’s data tool CrowdTangle to capture any posts on public pages, groups, or verified profiles on Facebook that linked to any one of these domains, resulting in 42.6 million posts from some 918,000 spaces between 1 January 2016 and 31 March 2021.

We’ve already mapped the network between Facebook spaces and the domains they shared, and this produces a clearly structured network involving various clusters around conservative and progressive politics in the US, conspiracist groups, non-English communities, and others. From here, though, we also filtered down our dataset to the 500 groups and 500 pages that were most prominent in our dataset (based on subscriber count, sharing activity, and user engagement), and then captured all of their Facebook posts since 2016, whether they contained links to our initial FakeNIX list of problematic domains or not. This produced some 70 million posts in total from some 954 spaces that were still accessible..

We then applied computational topic modelling to this dataset, and especially found wellness and conspirituality themes with problematic borrowings from African and Indigenous traditions; natural food and recipes themes; Catholic and Evangelical themes; entertainment themes; US conservative and progressive politics; and discussions of Israel and Palestine. US politics tends to dominate, in part due to the US focus of the initial FakeNIX list.

Mapping the 954 spaces by their topic similarity shows the interconnections between them, and a longitudinal analysis also clearly shows the impact of COVID-19 in connecting previously less connected groups – the wellness spaces are forming associations with the conspiracists and political extremists, for instance.