Toxicidade em Chats de Live Streaming: Um Estudo Comparativo com Coleta Síncrona em Twitch, YouTube e Kick
Resumo
Este artigo analisa padrões de toxicidade em chats de live streaming por meio da mineração de 6,8 milhões de mensagens coletadas simultaneamente na Twitch, YouTube e Kick durante 90 dias. As mensagens em português brasileiro foram processadas por um pipeline de PLN em duas etapas com modelos BERT, combinando polaridade semântica e categorização de toxicidade. A Kick apresentou a maior proporção de mensagens nocivas (21,4%), seguida pelo YouTube (19,6%) e pela Twitch (16,3%), com predominância de insultos e linguagem obscena. A análise temporal identificou picos recorrentes entre 11h e 16h (UTC-3), mostrando diferenças consistentes no perfil de toxicidade entre plataformas e indicando que a coleta síncrona é adequada para comparações multiplataforma em larga escala.Referências
Dreier, L. and Pirker, J. (2023). Toxicity in twitch live stream chats: Towards understanding the impact of gender, size of community and game genre. In 2023 IEEE Conference on Games (CoG), pages 1–4. IEEE.
Li, Y., Wang, C., and Liu, J. (2020). A systematic review of literature on user behavior in video game live streaming. International Journal of Environmental Research and Public Health, 17(9):3328.
Lobo, J. V. C., Barbosa, D. M., and de Freitas Melo, P. (2025). Investigando a dinâmica da propagação do ódio em cascatas de toxicidade nos chats ao vivo da twitch. In Brazilian Symposium on Multimedia and the Web (WebMedia), pages 331–339. SBC.
Morrier, J., Mahmassani, A., and Alvarez, R. M. (2025). Uncovering the viral nature of toxicity in competitive online video games. IEEE Transactions on Games.
Pérez, J. M., Giudici, J. C., and Luque, F. (2021). pysentimiento: A python toolkit for sentiment analysis and socialnlp tasks.
Ringer, C., Nicolaou, M., and Walker, J. (2020). Twitchchat: A dataset for exploring livestream chat. In Proceedings of the AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment, volume 16, pages 259–265.
Sheth, A., Shalin, V. L., and Kursuncu, U. (2022). Defining and detecting toxicity on social media: context and knowledge are key. Neurocomputing, 490:312–318.
Singh, A. K., Ghafouri, V., Such, J., and Suarez-Tangil, G. (2024). Differences in the toxic language of cross-platform communities. In Proceedings of the International AAAI Conference on Web and Social Media, volume 18, pages 1463–1476.
Tereshchenko, Y. and Hämäläinen, M. K. (2025). Efficient toxicity detection in gaming chats: A comparative study of embeddings, fine-tuned transformers and llms. Journal of Data Mining & Digital Humanities.
Trajano, D. (2023). toxicity-type-detection: A model for detecting types of toxicity in portuguese. [link]. Hugging Face model repository, accessed: 2026-02-22.
Vargas, F. A., Carvalho, I., de Góes, F. R., Benevenuto, F., and Pardo, T. A. S. (2021). Hatebr: A large expert annotated corpus of brazilian instagram comments for offensive language and hate speech detection. arXiv preprint arXiv:2103.14972.
Li, Y., Wang, C., and Liu, J. (2020). A systematic review of literature on user behavior in video game live streaming. International Journal of Environmental Research and Public Health, 17(9):3328.
Lobo, J. V. C., Barbosa, D. M., and de Freitas Melo, P. (2025). Investigando a dinâmica da propagação do ódio em cascatas de toxicidade nos chats ao vivo da twitch. In Brazilian Symposium on Multimedia and the Web (WebMedia), pages 331–339. SBC.
Morrier, J., Mahmassani, A., and Alvarez, R. M. (2025). Uncovering the viral nature of toxicity in competitive online video games. IEEE Transactions on Games.
Pérez, J. M., Giudici, J. C., and Luque, F. (2021). pysentimiento: A python toolkit for sentiment analysis and socialnlp tasks.
Ringer, C., Nicolaou, M., and Walker, J. (2020). Twitchchat: A dataset for exploring livestream chat. In Proceedings of the AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment, volume 16, pages 259–265.
Sheth, A., Shalin, V. L., and Kursuncu, U. (2022). Defining and detecting toxicity on social media: context and knowledge are key. Neurocomputing, 490:312–318.
Singh, A. K., Ghafouri, V., Such, J., and Suarez-Tangil, G. (2024). Differences in the toxic language of cross-platform communities. In Proceedings of the International AAAI Conference on Web and Social Media, volume 18, pages 1463–1476.
Tereshchenko, Y. and Hämäläinen, M. K. (2025). Efficient toxicity detection in gaming chats: A comparative study of embeddings, fine-tuned transformers and llms. Journal of Data Mining & Digital Humanities.
Trajano, D. (2023). toxicity-type-detection: A model for detecting types of toxicity in portuguese. [link]. Hugging Face model repository, accessed: 2026-02-22.
Vargas, F. A., Carvalho, I., de Góes, F. R., Benevenuto, F., and Pardo, T. A. S. (2021). Hatebr: A large expert annotated corpus of brazilian instagram comments for offensive language and hate speech detection. arXiv preprint arXiv:2103.14972.
Publicado
19/07/2026
Como Citar
CARDOSO, Rodrigo; SILVA, Carlos A..
Toxicidade em Chats de Live Streaming: Um Estudo Comparativo com Coleta Síncrona em Twitch, YouTube e Kick. In: BRAZILIAN WORKSHOP ON SOCIAL NETWORK ANALYSIS AND MINING (BRASNAM), 15. , 2026, Gramado/RS.
Anais [...].
Porto Alegre: Sociedade Brasileira de Computação,
2026
.
p. 215-226.
ISSN 2595-6094.
DOI: https://doi.org/10.5753/brasnam.2026.21938.
