Comparison of algorithms for detecting social bots in the 2018 Brazilian presidential elections using user characteristics

Authors

  • Bianca Lima Santos Universidade de São Paulo
  • Gabriel Estavaringo Ferreira Universidade de São Paulo
  • Marcelo Torres do Ó Universidade de São Paulo
  • Rafael Rodrigues Braz Universidade de São Paulo
  • Luciano Antonio Digiampietri Universidade de São Paulo https://orcid.org/0000-0003-4890-1548

DOI:

https://doi.org/10.5335/rbca.v13i1.11199

Keywords:

Bot, Classification, Social networks, Twitter, Elections, Machine learning

Abstract

The use of social bots for political purposes has become an increasingly relevant concern and has raised warningsabout the impact on democratic discussions. This paper presents a case study on the use of bots in politicaldiscussions during the second round of the 2018 Brazilian elections, aiming to build an automatic detection modelfor bots and comparing the use of explainable and non-explainable artificial intelligence algorithms. First, adataset was built by manually labeling accounts as bots or humans. Then linear regression algorithms, randomtrees, naive Bayesian, multilayer perceptron, and random forest were applied. It was identified that even simpleand explainable algorithms, such as random tree, perform similarly to more complex algorithms such as randomforest. Using only user’s characteristics, it was possible to identify more than 46% of the bots, but all modelsshowed a precision not greater than 52% in this task.

Downloads

Download data is not yet available.

Published

2020-11-09

Issue

Section

Original Paper

How to Cite

[1]
2020. Comparison of algorithms for detecting social bots in the 2018 Brazilian presidential elections using user characteristics. Brazilian Journal of Applied Computing. 13, 1 (Nov. 2020), 53–64. DOI:https://doi.org/10.5335/rbca.v13i1.11199.