Analyzing User Ideologies and Shared News During the 2019 Argentinian Elections

TL;DR

The study tackles how users' political ideologies influence the news they share on social media during Argentina's 2019 elections. It combines user-ideology classification from tweet content with scraping of news articles linked in those tweets, followed by sentiment bias and topic analyses using TF-IDF and Non-negative Matrix Factorization, to relate content bias and topics to user ideology. Key findings show that users predominantly share news biased toward their own coalitions, with clear cherry-picking effects across outlets, and that topic interests differ by ideology (e.g., Wage/Inflation favoring CL, Justice favoring CR). The work introduces a reusable analytical framework for measuring media and partisan agendas in polarized environments, offering a tool for cross-country comparisons and informing discussions on information diffusion and democratic processes.

Abstract

The extensive data generated on social media platforms allow us to gain insights over trending topics and public opinions. Additionally, it offers a window into user behavior, including their content engagement and news sharing habits. In this study, we analyze the relationship between users' political ideologies and the news they share during Argentina's 2019 election period. Our findings reveal that users predominantly share news that aligns with their political beliefs, despite accessing media outlets with diverse political leanings. Moreover, we observe a consistent pattern of users sharing articles related to topics biased to their preferred candidates, highlighting a deeper level of political alignment in online discussions. We believe that this systematic analysis framework can be applied to similar scenarios in different countries, especially those marked by significant political polarization, akin to Argentina.

Figures (7)

• Figure 1: Methodology pipeline.Top: example of raw data of tweets from social media Twitter (now X). In the left: hashtags were utilized to train a logistic regression model for classifying tweets as supportive of one candidate or the other. Users are assigned to the candidate for whom they exhibit the highest number of supportive tweets (see more details in zhou2021). In the right: the news URLs in the tweets are utilized to extract the text by web scrapping to execute then all the necessary steps leading to perform sentiment and topic analysis.
• Figure 2: Interpretation of the SB. This figure displays the probability of an article being favorable towards CL, CR, or neutral, given the value of SB measured over the article content. The shaded regions represent the 90% confidence intervals calculated by bootstrapping.
• Figure 3: Distribution of News Articles by Media Outlet.(A) Shown in gray, the distribution of unique news articles circulating on Twitter, categorized by their source media outlet (unique indicates that the frequency of news sharing is not considered). (B) Illustrated in yellow, the distribution of news articles shared by users, with each instance of news sharing counted independently, regardless of previous shares by others (in gray, the distribution from panel A for comparison). (C) The percentage distribution of news articles shared by media outlets among CL (in blue) and CR (in red) supporters, based on their sharing patterns.
• Figure 4: Mean Sentiment Bias ($\Bar{SB}$) by media outlet.$\Bar{SB}$ represents the average SB across all articles from a specific media outlet. Media outlets positioned on the left side are interpreted as having a bias towards the Center-Left (CL), while those on the right side are considered to have a bias towards the Center-Right (CR). Gray bars indicate the centered $99\%$ quantile of the estimator determined through bootstrapping, and stars denote those estimates significantly different from zero.
• Figure 5: (A) Cumulative Distribution of Sentiment Bias (SB) for shared news articles. Inset: average Sentiment Bias for news articles shared by CL and CR Partisans. The distributions illustrate the variability in the estimates calculated through bootstrapping. (B) Average Sentiment Bias for news articles shared by CL and CR partisans across media outlets. Each point represents the average sentiment bias across all articles from a given media outlet shared by each user group, with CL in blue and CR in red. Only media outlets with at least 100 articles shared by each group are included (see Supplementary Information). The horizontal bars indicate the centered $99\%$ quantile of the estimate obtained via bootstrapping. Black stars highlight instances where the difference in $\bar{SB}$ between CL and CR supporters is statistically significant, with $p < 0.01$.