This study explores the application of large language models (LLMs) to analyze political communication on social media during the 2022 Italian general elections. Leveraging OpenAI’s ChatGPT and the text-embedding-ada-002 model, we conducted an exploratory analysis to identify the themes in Facebook posts from 12 major Italian news outlets. These posts were collected in the three months leading up to the election. By clustering these posts and mapping their political parallelism, we examined the extent to which the ideological alignment of media sources influences the coverage of political topics. We employed a novel methodology combining supervised fine-tuning of an LLM for classifying political content and unsupervised clustering for thematic analysis. Our approach included the use of the Multi-Party Media Partisanship Attention Score (MP-MPAS) to quantify the partisan attention received by different news sources on Twitter. The integration of these methods allowed us to address two primary research questions: whether political parallelism is evident on social media and whether it extends to the topics covered. The results reveal significant political parallelism in the Italian media system, observable in the alignment of news sources with specific political parties and the distribution of themes across these sources. This study demonstrates the potential of LLMs to enhance the analysis of political communication, offering insights into the dynamics of media coverage and ideological biases during electoral campaigns.
Analisi Computazionale del Parallelismo Politico in Italia: Il Caso delle Elezioni 2022
Giglietto F.
Writing – Original Draft Preparation
;Righetti N.Writing – Original Draft Preparation
;Stanziano A.Writing – Original Draft Preparation
2024
Abstract
This study explores the application of large language models (LLMs) to analyze political communication on social media during the 2022 Italian general elections. Leveraging OpenAI’s ChatGPT and the text-embedding-ada-002 model, we conducted an exploratory analysis to identify the themes in Facebook posts from 12 major Italian news outlets. These posts were collected in the three months leading up to the election. By clustering these posts and mapping their political parallelism, we examined the extent to which the ideological alignment of media sources influences the coverage of political topics. We employed a novel methodology combining supervised fine-tuning of an LLM for classifying political content and unsupervised clustering for thematic analysis. Our approach included the use of the Multi-Party Media Partisanship Attention Score (MP-MPAS) to quantify the partisan attention received by different news sources on Twitter. The integration of these methods allowed us to address two primary research questions: whether political parallelism is evident on social media and whether it extends to the topics covered. The results reveal significant political parallelism in the Italian media system, observable in the alignment of news sources with specific political parties and the distribution of themes across these sources. This study demonstrates the potential of LLMs to enhance the analysis of political communication, offering insights into the dynamics of media coverage and ideological biases during electoral campaigns.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
