The second ICA 2018 session this morning is the one I have a paper in as well – we’re discussing the (scant) empirical evidence for echo chambers and filter bubbles. We start, though, with a paper by Anja Bechmann that is working with a broad sample of newsfeed data from Danish Facebook users.
The newsfeed potentially acts as a shared public news platform where people meet around shared news content. ‘News’ here might mean many things – journalistic, political news in a narrow sense, but also user-oriented relation news and the user updates of many kinds that newsfeed algorithms treat as news. The present paper also moves away from opinion-driven, attitudinal questions that look for partisan echo chambers or filter bubbles (which in their dyadic nature may work in the U.S., but not so much in multi-party Denmark), but rather looks to non-overlapping segments that share specific content.
So there are two potential trajectories here: a focus on source diversity (as measured by the diversity of outlets, programme sources, and URLs being shared); and a focus on content diversity (as measured by content formats, target demographics, and content viewpoints). Further, there is exposure diversity, which examines what audiences are reached and which users engaged with such content.
The dataset that this analysis is applied to is a broad-range sample of the Danish Facebook population. Users have volunteered to provide their data to the project; the userbase for the project is slightly more highly educated and slightly older than the Danish average, and not entirely representative of the overall Danish population.
For this group, the project has assessed the link similarity and semantic similarity of the content appearing in their newsfeeds. A heatmap of the link similarities shows that only some 10% of the entire sample population appears to be in bubbles that are separate from the mainstream community; this also changes for different threshold definitions of what we understand by ‘filter bubbles’, however. In the semantic analysis, some 28% are in outsider bubbles, but there is a limited overlap between these two filter bubble definitions. Sociodemographic factors did not seem to determine such filter bubble membership.
There is a significant need to better define what we even mean by a vague term such as ‘filter bubble’, then, and to operationalise that definition for empirical analysis. It is also important to note that such patterns change over time, e.g. over the course of the day, month, or year.
