Aprendizagem de máquina aplicada para identificação de preconceitos em textos de redes sociais

Abstract

Social networks are powerful tools capable of sharing, in seconds, diverse information with millions of users. This power can be dangerous when the disseminated information is hate speech, harming individuals and groups directly and indirectly. Systems that feed on texts and information from social networks are subject to collecting this type of data, data that can harm possible analyzes made in these systems that are based on texts. Depending on the objective, the algorithms can make erroneous decisions, directly affecting one or more individuals, a resolution that could be different if texts full of intolerance were not in the analyzed dataset, impairing the program’s judgment. Using methods of classification and supervised machine learning, this work sought to identify texts present in social networks in Portuguese that contain some type of discrimination or prejudice, contributing both to the end user of the social network and to developers who seek to identify and remove this type of content from your database.