An Integrated Approach for Detecting Hate Speech on Social Media Using Text and Emoji Vectorization

• Arthur Lima de Araújo Miranda UPE
• Cleyton Mário de Oliveira Rodrigues UPE

DOI: https://doi.org/10.5753/wics.2025.8136

Abstract

This paper proposes an integrated approach for hate speech detection in social media, combining three main dimensions: (1) fusion of Brazilian Portuguese datasets (HateBR and TuPy-E), (2) joint processing of texts and emojis, and (3) a two-stage classification architecture (binary and multiclass). Using the BERTimbau model adapted to capture semantic relations and emoji representations, the system first performs binary classification (hate vs non-hate) followed by specific categorization (xenophobia, gender/sexuality, etc.). Results achieved 85% accuracy in the binary stage and up to 86% in specific categories. We discuss the relationship between data volume and performance, as well as future strategies for model improvement, including the use of LLMs (Large Language Models) and metadata integration.

References

Aizawa, A. (2003). An information-theoretic perspective of tf-idf measures. Information Processing Management, 39(1):45–65.

Almaliki, M., Almars, A. M., Gad, I., and Atlam, E.-S. (2023). Abmm: Arabic bert-mini model for hate-speech detection on social media. Electronics, 12(1048).

ANDES (2023). Brasil lidera discurso de ódio nas redes sociais contra população lgbtqiap+. Acesso em: 20 de março de 2025.

BR, U. (2023). O discurso de ódio ’é um dos sinais de alerta de genocídio e de outros crimes atrozes,’ alerta guterres. Acesso em: 20 de março de 2025.

BR, U. (2024). Como podemos ajudar a combater o discurso de ódio nas redes sociais. Acesso em: 20 de março de 2025.

Caseli, H. M. and (orgs.), M. G. V. N. (2023). Processamento de Linguagem Natural: Conceitos, Técnicas e Aplicações em Português. BPLN.

Ebecken, N., Lopes, M., and Costa, M. (2003). Mineração de textos. In Capítulo 13, p. 337–370. Manole.

Géron, A. (2017). Hands-On Machine Learning with Scikit-Learn and TensorFlow: Concepts, Tools, and Techniques to Build Intelligent Systems. O’Reilly Media, Inc., 1st edition.

Lorena, A. C. and de Carvalho, A. C. (2007). Uma introdução às support vector machines. Revista de Informática Teórica e Aplicada, 14(2):43–67.

Nacional, J. (2023). Denúncias de crimes envolvendo discurso de ódio nas redes sociais triplicaram nos últimos 6 anos, aponta levantamento. Acesso em: 20 de março de 2025.

Nascimento, F. R. S., Cavalcanti, G. D. C., and Costa-Abreu, M. D. (2023). Exploring automatic hate speech detection on social media: A focus on content-based analysis. SAGE Open, April-June 2023:1–19.

Oliveira, F., Reis, V., and Ebecken, N. (2023). Tupy-e: Detecting hate speech in brazilian portuguese social media with a novel dataset and comprehensive analysis of models. arXiv preprint.

Vargas, F., Carvalho, I., Goés, F., Pardo, T. A., and Benevenuto, F. (2022). Hatebr: A large expert annotated corpus of brazilian instagram comments for offensive language and hate speech detection. arXiv preprint.