Skip to content

A Transformers-based model that classifies the political bias of Brazilian Portuguese tweets as either conservative, liberal, or neutral.

Notifications You must be signed in to change notification settings

Daniel-EST/political-bias

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

10 Commits
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

political-bias

A Transformers-based model that classifies the political bias of Brazilian Portuguese tweets as either conservative, liberal, or neutral. This work was presented as part of a subject at system engineering and computer science and masters program.

Results:

Accuracy: 80% F1-Score: 75%

Methodology:

Identifying political biases on Twitter posts in Brazilian Portuguese

Due to the rise of microblogging platforms like Twitter, social media has evolved from a simple means of communication to a platform where users can share their views on various social and political issues. However, the abundance of biased information on social media can contribute to polarization. To address this issue, I trained a model to classify political bias on Twitter using BERTimbau, a Portuguese implementation of Google's BERT, and employing a Post-Based Method strategy.

The approach involved gathering posts from users who had already been categorized according to their political leanings. I assumed that all posts made by an account were inherently aligned with the same political bias as the account's categorization. It resulted in the collection of 243,624 tweets, with 95,630 classified as conservative, 22,233 as neutral, and 125,761 as liberal. However, it should be noted that not all tweets collected from an account are necessarily related to politics.

The approach yielded promising results, with an accuracy of 80% and an F1-Score of 75%. The outcomes suggest that the proposed approach could be useful in identifying political bias on Twitter in Portuguese. However, the findings need to be further validated, and the limitations of the approach should be taken into account. For example, the assumption that all posts by an account are aligned with the account's categorization may not always hold.

Examples:

Pre-processed tweet Prediction Actual
"O que há de antidemocrático em pedir o aprimoramento do processo eleitoral?" conservative conservative
"Por ser conservador, Filipe Martins está sendo perseguido" conservative conservative
"O trunfo de Bolsonaro para tentar crescer na cidade de São Paulo (índice da mão traseira apontando para baixo: tom de pele médio)" neutral neutral
"Meu Deus, que gente chata! Boa noite!" liberal liberal
"Bolsonaro fala AO VIVO na Globo que acabou com a mamata da Globo! rolando de rir no chão" neutral liberal
"Enquanto Lula louva, elogia e agradece aqueles que agridem opositores políticos para defendê-lo, o Presidente sempre deixou claro que dispensa esse tipo de apoio. Desde 2018. A diferença entre os dois é cristalina e não comporta nenhum tipo de falsa equivalência. liberal conservative
"(luz do carro da policia) FAMOSOS Julia Fox diz que só namorou Kanye West para que ele deixasse Kim Kardashian em paz." neutral liberal

About

A Transformers-based model that classifies the political bias of Brazilian Portuguese tweets as either conservative, liberal, or neutral.

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published