Online Information Consumption and Self-Reported Importance on Sociopolitical Issues: Investigating potential Dissonance using Large Language Models
Year: 2024
Type: dissertation
Study design: Observational Study
Abstract: The rapid growth of digital media and online platforms has transformed the way citizens access and consume socio-political information. Despite this democratization in the information dissemination, some researchers have suggested that digital media has also contributed to fragmentation and polarization of political opinions. So far, surveys on online information consumption patterns have not been able to adequately explain this phenomenon. This thesis aims to contribute to the investigation of dissonance between individuals’ self-reported importance on socio-political issues and their actual online media consumption patterns, focusing on the current conflict in Ukraine. Based on a novel methodological approach, the study combines unobtrusive digital behavioral data from a panel of German internet users with survey responses collected over a 12-month period. Using natural language processing techniques, in particular topic modeling and large language models, we identify and categorize sociopolitical issues present in online content consumed by individuals. Meta-Llama-3-8B-Instruct has proven to be the most efficient model for our purposes. These results will later be compared with those of surveys on the importance of various sociopolitical issues. Our findings reveal a prevalent misalignment, with underexposure to important issues being more common than overexposure. Contrary to some theories of passive media consumption, users rarely engage with more Ukraine-related content than they deem necessary. Furthermore, while political interest and education show some correlation with dissonance, these factors are not strongly predictive of alignment. Methodologically, this research also highlights the effectiveness of LLMs in analyzing and categorizing political content, opening new possibilities for more in-depth analysis of media consumption. Overall, it could be concluded that this thesis enhances our understanding of political engagement in the digital age and paves the way for future studies in computational social science.
Source: Aaltodoc (Aalto University)
Author: Marin Colome, Miquel
Language: English
Topic: Social Media and Politics
Subfield: Communication
Field: Social Sciences
Domain: Social Sciences
Open Access status: green
Funder: Aalto-Yliopisto
OpenAlex ID: W7014159467