And the final speaker in this final paper session at the AoIR 2025 conference is my QUT colleague Kate O’Connor-Farfan, who focusses on the 2022 Australian federal election. She begins by noting the tensions between scale and depth in social media analysis: computational methods often privilege scale over depth, and there are now attempts to overcome this with the use of LLMs.
Her work draws on data from the two leading candidates’ – Scott Morrison and Anthony Albanese – Facebook pages, from which she extracted the key narrative structures. An preliminary analysis of the key terms used by these political leaders shows their own narrative approaches, but there is also an opportunity to examine how commenters on these pages are referring to these politicians (and to other political actors).
Natural Language Processing tools provide a pathway to this, but their outputs are often complicated to parse, and need to be aggregated into broader discursive structures for further analysis. Individual sentences can contain multiple assessments of politicians, for instance, and the ways these politicians are referred to are also highly variable.
The tracing of narrative structures within the data can help with this: this is independent of the specific words used, and instead looks for particular narrative constructions – for instance, direct assessments made of political actors, or objects of desire articulated in the comments. The comparison of these patterns can then also point to polarisation between different political actors and groups.
Candidates themselves often attached positive assessments to themselves, and ignored their rivals, while the commenters themselves often responded with more critical direct assessments of these candidates; these can be classified both by their target, the nature of the assessment, and the positive or negative strength of the assessment, for instance.
Both candidates were criticised for the relationship with China, for instance; Albanese was praised for his moral character, and criticised for his weakness and experience; Morrison was sometimes praised for his experience, and often criticised for his lack of morals, his character, and his belligerence. Such assessments were more aggressive in comments on Albanese’s posts than in comments on Morrison’s posts.
The moral dimension of these assessments (rather than factual judgments) points especially to an affective dimension of polarisation between the candidates. Overall, though, the methodological pathway also points to the opportunity to use computational analysis for more complex studies.
