Methods for Understanding Cumulative Public Opinion Formation in Social Media

The next session at IAMCR 2024 starts with Svetlana Bodrunova, who introduces a methodological focus in the study of topic evolution in user talk on social media platforms. Key to this is the use of artificial intelligence tools.

Deliberative public communication research tends to remain strongly influenced by Habermasian normativity, but this is not necessarily very productive. It ignores the right of participants not to be deliberative, and therefore fails to fully understand the phenomenon of dissonant public spheres, or the cumulative nature of public discussion. We need to better understand how opinions accumulate online.

Big data approaches are central to detecting this accumulation of opinions; this includes quantitative accumulation as well as qualitative evolution and mutation. The analysis of these processes can be aided by a combination of AI-based methods and topic modelling approaches. This resembles a data analysis tree which disassembles the overall dataset into various thematic and subthematic branches that can be understood through the use of such methods.

This is an open-ended process that does not require researchers to pre-define the topics that are expected to emerge from the dataset; rather, there is an iterative modelling process which involves text embeddings, dimensionality reduction, topic reduction into macro-topics, recursive topic modelling to identify bifurcation points, and the abstractive summarisation of topic branches. Ideally, this should also be able to work in the dynamic environment of a live social media platform, where new texts are constantly added to the existing dataset.

In

Interestingly, this summarisation resulted in AI tools generating rhetorical questions based on the data they processed – and this pointed strongly towards polarised debates where public opinion was not yet settled. This might be further operationalised in the study of polarisation.