Anatomia dos Repositórios de Dados para Análise e Detecção de Toxicidade em Português
Resumo
Online hate speech detection requires a rigorous examination of the datasets used to train models. We analyze six Brazilian Portuguese datasets annotated for hate speech or toxicity, combining HurtLex-based lexical profiling with cross-dataset evaluation (using BERTimbau embeddings and XGBoost) to assess how their lexical “anatomy” affects generalization. Although the datasets share a similar macro-level focus, they diverge in how specific terms are used and labeled across platforms. Broader, more heterogeneous toxic vocabularies yield better cross-domain performance, whereas narrow, profanity-centered labeling causes severe generalization gaps, even when lexical overlap is high. Warning! This work contains examples of offensive and hateful language.
Palavras-chave:
hate speech, offensive speech, toxicity, NLP, Portuguese, crossdomain generalization
Referências
Lora Aroyo, Lucas Dixon, Nithum Thain, Olivia Redfield, and Rachel Rosen. 2019. Crowdsourcing subjective tasks: The case study of understanding toxicity in online discussions. In WWW ’19 Companion. Association for Computing Machinery, Stroudsburg, PA, USA, 1100–1105.
Elisa Bassignana, Valerio Basile, and Viviana Patti. 2018. HurtLex: A multilingual lexicon of words to hurt. In CEUR Workshop Proc., Vol. 2253. 1–6.
Tianqi Chen and Carlos Guestrin. 2016. XGBoost: A scalable tree boosting system. In KDD. 785–794.
Rogers P. de Pelle and Viviane P. Moreira. 2017. Offensive comments in the Brazilian web: A dataset and baseline results. In BraSNAM.
Paula Fortuna, Joao Rocha da Silva, Leo Wanner, and Sérgio Nunes. 2019. A hierarchically-labeled Portuguese hate speech dataset. In ALW3 – ACL. 94–104.
Paula Fortuna, Juan Soler, and Leo Wanner. 2020. Toxic, hateful, offensive or abusive? What are we really classifying? An empirical analysis of hate speech datasets. In LREC. 6786–6794.
Samuel Guimarães, Gabriel Kakizaki, Philipe Melo, Márcio Silva, Fabricio Murai, Julio C. S. Reis, and Fabrício Benevenuto. 2023. Anatomy of hate speech datasets: Composition analysis and crossdataset classification. In ACM HT. 33:1–33:11.
Fatih Karabiber. 2025. Jaccard similarity. [link] Accessed: August 11, 2025.
Alfirna Rizqi Lahitani, Adhistya Erna Permanasari, and Noor Akhmad Setiawan. 2016. Cosine similarity to determine similarity measure: Study case in online essay assessment. In CITSM. 1–6.
João Augusto Leite, Diego Silva, Kalina Bontcheva, and Carolina Scarton. 2020. Toxic language detection in social media for Brazilian Portuguese: New dataset and multilingual analysis. In AACL-IJCNLP. 914–924.
Lorena Souza Moreira, Paula Teresa M. Gibrim, Leonardo Rocha, and Julio C. S. Reis. 2026. Anatomy of Data Repositories for the Analysis and Detection of Toxicity in Portuguese. In PROPOR, Vol. 1. 456–466.
Fabio Poletto, Valerio Basile, Manuela Sanguinetti, Cristina Bosco, and Viviana Patti. 2021. Resources and benchmark corpora for hate speech detection: A systematic review. Lang. Resour. Eval. 55, 3 (2021), 477–523.
Felipe Ramos de Oliveira, Victoria Dias Reis, and Nelson Francisco Favilla Ebecken. 2024. Detecting hate speech on Brazilian social media: New dataset and analysis. In CILAMCE.
Yael Ravin and Claudia Leacock. 2000. Polysemy: Theoretical and computational approaches. Oxford University Press.
Isadora Salles, Francielle Vargas, and Fabrício Benevenuto. 2025. HateBRXplain: A benchmark dataset with human-annotated rationales for explainable hate speech detection in Brazilian Portuguese. In COLING. 6659–6669.
Fábio Souza, Rodrigo Nogueira, and Roberto Lotufo. 2020. BERTimbau: Pretrained BERT models for Brazilian Portuguese. In BRACIS. 403–417.
Douglas Trajano, Rafael H. Bordini, and Renata Vieira. 2024. OLID-BR: Offensive language identification dataset for Brazilian Portuguese. Lang. Resour. Eval. 58, 4 (2024), 1263–1289.
Francielle Vargas, Isabelle Carvalho, Thiago A. S. Pardo, and Fabrício Benevenuto. 2025. Context-aware and expert data resources for Brazilian Portuguese hate speech detection. Nat. Lang. Process. 31 (2025), 435–456.
Zeerak Waseem and Dirk Hovy. 2016. Hateful symbols or hateful people? Predictive features for hate speech detection on Twitter. In NAACL SRW. 88–93.
Elisa Bassignana, Valerio Basile, and Viviana Patti. 2018. HurtLex: A multilingual lexicon of words to hurt. In CEUR Workshop Proc., Vol. 2253. 1–6.
Tianqi Chen and Carlos Guestrin. 2016. XGBoost: A scalable tree boosting system. In KDD. 785–794.
Rogers P. de Pelle and Viviane P. Moreira. 2017. Offensive comments in the Brazilian web: A dataset and baseline results. In BraSNAM.
Paula Fortuna, Joao Rocha da Silva, Leo Wanner, and Sérgio Nunes. 2019. A hierarchically-labeled Portuguese hate speech dataset. In ALW3 – ACL. 94–104.
Paula Fortuna, Juan Soler, and Leo Wanner. 2020. Toxic, hateful, offensive or abusive? What are we really classifying? An empirical analysis of hate speech datasets. In LREC. 6786–6794.
Samuel Guimarães, Gabriel Kakizaki, Philipe Melo, Márcio Silva, Fabricio Murai, Julio C. S. Reis, and Fabrício Benevenuto. 2023. Anatomy of hate speech datasets: Composition analysis and crossdataset classification. In ACM HT. 33:1–33:11.
Fatih Karabiber. 2025. Jaccard similarity. [link] Accessed: August 11, 2025.
Alfirna Rizqi Lahitani, Adhistya Erna Permanasari, and Noor Akhmad Setiawan. 2016. Cosine similarity to determine similarity measure: Study case in online essay assessment. In CITSM. 1–6.
João Augusto Leite, Diego Silva, Kalina Bontcheva, and Carolina Scarton. 2020. Toxic language detection in social media for Brazilian Portuguese: New dataset and multilingual analysis. In AACL-IJCNLP. 914–924.
Lorena Souza Moreira, Paula Teresa M. Gibrim, Leonardo Rocha, and Julio C. S. Reis. 2026. Anatomy of Data Repositories for the Analysis and Detection of Toxicity in Portuguese. In PROPOR, Vol. 1. 456–466.
Fabio Poletto, Valerio Basile, Manuela Sanguinetti, Cristina Bosco, and Viviana Patti. 2021. Resources and benchmark corpora for hate speech detection: A systematic review. Lang. Resour. Eval. 55, 3 (2021), 477–523.
Felipe Ramos de Oliveira, Victoria Dias Reis, and Nelson Francisco Favilla Ebecken. 2024. Detecting hate speech on Brazilian social media: New dataset and analysis. In CILAMCE.
Yael Ravin and Claudia Leacock. 2000. Polysemy: Theoretical and computational approaches. Oxford University Press.
Isadora Salles, Francielle Vargas, and Fabrício Benevenuto. 2025. HateBRXplain: A benchmark dataset with human-annotated rationales for explainable hate speech detection in Brazilian Portuguese. In COLING. 6659–6669.
Fábio Souza, Rodrigo Nogueira, and Roberto Lotufo. 2020. BERTimbau: Pretrained BERT models for Brazilian Portuguese. In BRACIS. 403–417.
Douglas Trajano, Rafael H. Bordini, and Renata Vieira. 2024. OLID-BR: Offensive language identification dataset for Brazilian Portuguese. Lang. Resour. Eval. 58, 4 (2024), 1263–1289.
Francielle Vargas, Isabelle Carvalho, Thiago A. S. Pardo, and Fabrício Benevenuto. 2025. Context-aware and expert data resources for Brazilian Portuguese hate speech detection. Nat. Lang. Process. 31 (2025), 435–456.
Zeerak Waseem and Dirk Hovy. 2016. Hateful symbols or hateful people? Predictive features for hate speech detection on Twitter. In NAACL SRW. 88–93.
Publicado
27/10/2026
Como Citar
MOREIRA, Lorena Souza; GIBRIM, Paula Teresa M.; ROCHA, Leonardo; REIS, Julio C. S..
Anatomia dos Repositórios de Dados para Análise e Detecção de Toxicidade em Português. In: CONCURSO DE TRABALHOS DE INICIAÇÃO CIENTÍFICA - SIMPÓSIO BRASILEIRO DE SISTEMAS MULTIMÍDIA E WEB (WEBMEDIA), 32. , 2026, Lavras/MG.
Anais [...].
Porto Alegre: Sociedade Brasileira de Computação,
2026
.
p. 46-49.
ISSN 2596-1683.
DOI: https://doi.org/10.5753/webmedia_estendido.2026.34472.
