My own presentation, on behalf of a much larger team at the QUT Digital Media Research Centre, concluded the final panel at the DigiWorld network ECREA preconference on election campaigning in Brno. For this, we set ourselves an ambitious task: first, given the increasing fragmentation of social media activity across a growing number of platforms, we wanted to explore how we might conduct a true cross-platform analysis of public discussions around the 2025 Australian election campaign, using data from platforms including Facebook, Instagram, Reddit, and YouTube.
Second, because this meant using the Meta Content Library for access to candidates' and campaigners' posts, and user comments in response to them, on Facebook and Instagram, it meant doing the analysis and visualisation of such patterns within the MCL, since due to Meta's 'data clean room' framework (which Laura Vodden and I cover in our new article in Political Communication Report) it is virtually impossible to export anything other than the aggregate graphs derived from these data from the MCL. This severely restricts the analytical tools and packages we have at our disposal.
Third, the Meta Content Library automatically wipes all data accessed by the researcher on the first of each month, for reasons of 'privacy'. This meant that we needed to gather data, process the dataset, consolidate results into a cross-platform analysis, and visualise the results within a single month, developing all approaches for this (at least this first time around) absolutely from scratch. This meant approaching this as a month-long data sprint, which we did in August 2026.
We had reason to be hopeful that this would work, though, as we could build on the principles and approaches of the practice mapping method we have introduced in our recent work (published in Social Media + Society and Network Science). While the four platforms are necessarily quite distinct from each other, we could nonetheless identify several practices which are universally present across these (and other) platforms: most simply, the use of language for communication (which even without being able to use word embeddings or other advanced techniques that are unavailable in the MCL we could approach using simpler but powerful methods such as TF-IDF), and the referencing of others – individuals, organisations, locations, etc. – in such communication (which we can assess using standard Named Entity Recognition techniques). There are more such platform-transcendent practices, but we focussed on these two for now.
Applying TF-IDF and NER to the collected posts from each account in any of the datasets produced two vectors describing the account's communicative practices during the election period: their overall communicative focus (which may or may not address specific themes or topics of particular interest), and their attention to particular selections of entities (which may also indicate particular affinities or antipathies). Comparing these vectors for all pairs of accounts, across all platforms, enabled us to generate a cross-platform matrix (and ultimately, network) of communicative similarities between accounts, and to explore how these cluster together around particular combinations of communicative practices. We could then read these as representing specific topics, attitudes, and styles of communication.
So, for instance, mainstream politicians' accounts largely clustered together not because they represented the same parties, but because they communicated (mostly on Facebook and Instagram) like politicians – and thus very differently from ordinary, 'normal' people. Those other users – Facebook and Instagram commenters, Reddit posters, some YouTube channels – posted very differently: they covered different themes, did so with different attitudes, and expressed themselves in different styles. This ranged from what looks like genuine foreign politics and policy discussions, especially on Reddit, to blunt characterisations of politicians as liars and even outright populist and conspiracist propaganda, especially in Facebook comments, with Instagram somewhere in the middle.
This first, month-long data sprint could only deliver the proof of concept for this approach: we needed to develop our data gathering, processing, analysis, and visualisation approaches and scripts first, and that's taken us the whole month of August. Next time around, with those scripts now in hand, we should be able to do all this much more quickly, leaving more time for in-depth analysis and enabling us to explore more carefully what the various clusters in our practice mapping network represent. But these first results are already intriguing, suggesting as they do that different platforms play very different (but at least partially overlapping) roles in public debate, and that how ordinary voters discuss an election diverges widely from the polished rhetoric employed by most mainstream parties (but not as much from that of the far-right populists).
We'll return to this work a little further down the track, once we've recovered from this initial data sprint and reviewed its outcomes some more. We're also hoping to share some of the scripts we developed, so others can replicate this cross-platform practice mapping approach for themselves. For now, though, here are the slides for the current presentation:
