Analysing Filter Bubbles in the Facebook Newsfeed

The next presenter at Future of Journalism 2017 is Anja Bechmann, who shifts our focus to news engagement within the private and semi-private spaces of Facebook. Here, the Facebook newsfeed serves at least in part also as a news platform, where news stories are shared and curated in a collaborative fashion. News, here, is variously a journalistically, user-, and algorithmically defined concept.

The investigation of the newsfeed can also help to detect 'filter bubbles', defined as non-overlapping content segments. Key questions here address source diversity, content diversity, and exposure diversity, as experienced by Facebook users; this can be addressed both by analysing link sharing patterns (through measuring link similarity across users) and observing the shared semantic spaces in which newsfeed updates operate (using LDA and related semantic analysis tools).

Link overlaps shows that less than 10% of all accounts exist in a distinct filter bubble that does not overlap with other communities; other communities do emerge, but overlap considerably with each other, too. An analysis of the semantic spaces inhabited by each newsfeed shows that the vast majority of personal newsfeeds studied here form one large shared semantic space, while all others exist in highly idiosyncratic spaces instead.

Education, gender, residence, or age have no effect here; instead, the main driver is a user's level of sociality (measured by friend numbers, group memberships, page likes), with users with lower sociality more likely to be placed outside the majority semantic space.