Análise da percepção sobre o viés reproduzido pelos algoritmos de reconhecimento facial

Abstract

The advancement in algorithm development enables the execution of tasks primarily performed by humans, easily replicable by Artificial Intelligence (AI) systems. Facial recognition systems, for instance, are adopted by companies and organizations to conduct studies with the purpose of automating tasks. In this process, problems arise concerning the results produced by these systems. These problems can be caused by algorithmic bias. In light of this, this study conducts an analysis of the perception of bias in the comments extracted from two databases on the social media network Reddit, where users commented on posts with facial recognition as the main topic. The methodology consisted of extracting and analyzing sentiments from the comments using the VADER sentiment analyzer from the Python Natural Language Toolkit (NLTK) and obtaining the most frequent terms for subsequent analysis. Positive and negative sentiments were found in similar quantities, and the extraction of the most mentioned expressions revealed the relationship between rights violation and the sentiments identified