Analyzing Social Media's Impact on Commemorative Narratives with a Large Language Model
- Jan Babnik (IRRIS Insitute)
- Jure Koražija (IRRIS Insitute)
Abstract
Cultural memories are shared memories, narratives, and representations of the past that are collectively held, transmitted, and preserved within a cultural group or society. They help shape a group's identity and provide a sense of continuity and connection across generations. We present a comprehensive method for conceptualizing and measuring the impact of social media on the transformation of cultural memories. The conceptualization is based on changes in patterns of deliberation, while measurements are conducted on textual analysis of social media using machine learning approaches.
Our hypothesis is that social media transform discussions on significant historical events by altering deliberative dynamics and amplifying the polarizing function of commemorative narratives, which ultimately undermines the coherence of the public sphere. To investigate how the deliberative patterns are being transformed, we developed a computational approach grounded in the theory of deliberative democracy. The specific theoretical model is Greimas' actantial model which is supported with natural language processing tools to extract relevant information from social media. We first automatically identify narratives associated with specific commemorations. These narratives are then analyzed in terms of legitimization in decision-making patterns specific to different democratic models (illiberal, liberal, epistocratic, participatory, deliberative, radical, and algorithmic) using large language models (LLMs). Next, we apply LLMs to identify tailored deliberative democracy indicators that measure the quality and goals of discussions on commemorations, emphasizing a summative over an additive approach in deliberative democracy studies.
The paper motivates the use of various language technologies and tasks, such as LLMs, topic modeling, and sentiment analysis, to automatically extract relevant information throughout the analysis. The approach is demonstrated using the Slovenian commemoration of the Day of Resistance as a case study. The results show the divisive nature of this commemoration and the transformations of cultural memories associated with it in social media relative to more established sources, such as newspapers and books.
We underscore the universal applicability of the demonstrated approach, including the potential to develop an automated pipeline for analyzing debates on divisive collective memories in various national and international contexts. Special attention is given to identifying points of conflict and establishing areas of common ground through computational analysis.