Arte ou Abuso? Uma Abordagem Comparativa para Classificação de ASCII Art NSFW na Steam
Resumo
ASCII Art is a technique that uses monospaced characters to form images, standing at the boundary between text and visual representation. Because it depends on the spatial arrangement of characters, this type of content is often processed as text by automated systems, even though it is interpreted visually by humans, which makes it difficult to detect using traditional moderation mechanisms based on Natural Language Processing (NLP). In this context, ASCII representations have been used to disseminate NSFW (Not Safe For Work) content, that is, material inappropriate for work or public environments, including nudity, explicit sexuality, graphic violence, and offensive language. Therefore, this study presents a comparative analysis for detecting NSFW content in ASCII Art using messages extracted from the Steam platform, one of the largest digital communities focused on video games and characterized by a high volume of user interactions. Vision-based architectures built on Convolutional Neural Networks were compared with the Longformer language model and with zero-shot and few-shot Qwen2.5-7B baselines, in both text and vision-language variants. From approximately 100 million messages, around 62,000 ASCII Art representations were identified, which underwent duplicate removal and the incorporation of samples from other sources, resulting in the final corpus used in this paper. Under stratified cross-validation with a corrected paired 𝑡-test, the three CNN architectures achieved statistically equivalent Macro-F1, significantly outperforming Longformer and the Qwen2.5-7B baselines. These results confirm the superiority of visual approaches over the textual paradigm, emphasizing the importance of preserving the two-dimensional spatial structure of ASCII Art. Warning: this work and the referenced data contain examples of potentially sensitive content, including material that may cause discomfort or contain sexual
Palavras-chave:
ASCII art, NSFW detection, content moderation, CNN, zeroshot, LLM, VLM
Referências
Iz Beltagy, Matthew E Peters, and Arman Cohan. 2020. Longformer: The long-document transformer. arXiv preprint arXiv:2004.05150 (2020).
Sergey Berezin, Reza Farahbakhsh, and Noel Crespi. 2024. Read over the lines: Attacking llms and toxicity detection systems with ascii art to mask profanity. arXiv preprint arXiv:2409.18708 (2024).
Daniel Bicho, Artur Ferreira, and Nuno Datia. 2020. A Deep Learning Approach to Identify Not Suitable for Work Images. i-ETC: ISEL Academic Journal of Electronics, Telecommunications and Computers 6, 1 (2020), 1–9. DOI: 10.22266/ijies2022.0228.19
Falcon Dai and Zheng Cai. 2017. Glyph-aware Embedding of Chinese Characters. In Proceedings of the First Workshop on Subword and Character Level Models in NLP (SCLeM). Association for Computational Linguistics, 64–69.
Mustafa Eriş and Mustafa Kaya. 2023. Leveraging Deep Learning for Identification of Illicit Images in Digital Forensic Investigations. Traitement du Signal 40, 6 (December 2023), 2539–2552. DOI: 10.18280/ts.400617
Joseph L Fleiss. 1971. Measuring nominal scale agreement among many raters. Psychological bulletin 76, 5 (1971), 378.
Akira Fujisawa, Kazuyuki Matsumoto, Kazuki Ohta, Minoru Yoshida, and Kenji Kita. 2020. Ascii art classification model by transfer learning and data augmentation. In Fuzzy Systems and Data Mining VI: Proceedings of FSDM 2020. SAGE Publications 1 Oliver’s Yard, 55 City Road, London, EC1Y 1SP, 608–618.
Paula Gibrim, Vinicius Lodi, Gabriel Andrade, Marcus Ribeiro, Mauricio Gruppi, Daniel Barbosa, Philipe Melo, and Julio C. S. Reis. 2026. Toxic Steam: A Large-Scale Characterization and Prediction of Toxic Users in a Gaming Platform. In Proceedings of the AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment (AIIDE). In press.
Paula Gibrim, Marcus Vinicius Guerra Ribeiro, Lorena Moreira, Daniel Mendes Barbosa, Philipe Melo, and Julio C. S. Reis. 2026. O Fenômeno da Toxicidade em Jogos Digitais: Uma Caracterização da Comunidade Brasileira na Steam. In Anais do XXV Simpósio Brasileiro de Jogos e Entretenimento Digital (SBGames). SBC, Porto Alegre, RS, Brasil. In press.
Paul Hardy and Roman Czyborra. 2023. GNU Unifont (Version 15.0.04). [link]. Acessado em: 28 jun. 2026.
Devamardeep Hayatpur, Brian Hempel, Kathy Chen, William Duan, Philip Guo, and Haijun Xia. 2024. Taking ascii drawings seriously: How programmers diagram code. In Proceedings of the 2024 CHI Conference on Human Factors in Computing Systems. 1–16.
Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. 2015. Deep Residual Learning for Image Recognition. arXiv:1512.03385 [cs.CV] [link]
Sture Holm. 1979. A Simple Sequentially Rejective Multiple Test Procedure. Scandinavian Journal of Statistics 6, 2 (1979), 65–70.
Andrew Howard, Mark Sandler, Grace Chu, Liang-Chieh Chen, Bo Chen, Mingxing Tan, Weijun Wang, Yukun Zhu, Ruoming Pang, Vijay Vasudevan, Quoc V Le, and Hartwig Adam. 2019. Searching for MobileNetV3. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV). 1314–1324.
Shalini Jangra, Zaid Almahmoud, Suparna De, Gareth Tyson, Ehsan Ul Haq, and Nishanth Sastry. 2026. Understanding the Complexities of Responsibly Sharing NSFW Content Online. In Proceedings of the International AAAI Conference on Web and Social Media, Vol. 20. 1083–1096.
Qi Jia, Xiang Yue, Shanshan Huang, Ziheng Qin, Yizhu Liu, Bill Yuchen Lin, and Yang You. 2024. Visual perception in text strings. arXiv preprint arXiv:2410.01733 (2024).
Fengqing Jiang, Zhangchen Xu, Luyao Niu, Zhen Xiang, Bhaskar Ramasubramanian, Bo Li, and Radha Poovendran. 2024. Art-Prompt: ASCII Art-based Jailbreak Attacks against Aligned LLMs. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). Association for Computational Linguistics, Bangkok, Thailand, 15157–15173. DOI: 10.18653/v1/2024.acl-long.809
J. R. Landis and G. G. Koch. 1977. The measurement of observer agreement for categorical data. Biometrics 33, 1 (1977), 159–174.
Tsung-Yi Lin, Priya Goyal, Ross Girshick, Kaiming He, and Piotr Dollár. 2017. Focal loss for dense object detection. In Proceedings of the IEEE international conference on computer vision. 2980–2988.
Ilya Loshchilov and Frank Hutter. 2017. Decoupled weight decay regularization. arXiv preprint arXiv:1711.05101 (2017).
Kerry Luo, Michael Fu, Joshua Peguero, Husnain Malik, Anvay Patil, Joyce Lin, Megan Van Overborg, Ryan Sarmiento, and Kevin Zhu. 2025. ASCIIBench: Evaluating Language-Model-Based Understanding of Visually-Oriented Text. arXiv preprint arXiv:2512.04125 (2025).
Claude Nadeau and Yoshua Bengio. 2003. Inference for the Generalization Error. Machine Learning 52, 3 (2003), 239–281. DOI: 10.1023/A:1024068626366
Qwen Team. 2025. Qwen2.5 Technical Report. arXiv preprint arXiv:2412.15115 (2025).
Qwen Team. 2025. Qwen2.5-VL Technical Report. arXiv preprint arXiv:2502.13923 (2025).
Marcus Vinicius Guerra Ribeiro, Bernardo Cunha, Vitor Andrade, Henrique Campos, Daniel Mendes Barbosa, and Philipe Melo. 2026. Livre para Quem? Uma Análise das Discrepâncias entre a Classificação Etária Indicativa e o Conteúdo Adulto Autodeclarado na Steam. In Anais do XXV Simpósio Brasileiro de Jogos e Entretenimento Digital (SBGames). SBC, Porto Alegre, RS, Brasil. In press.
Marcus Vinicius Guerra Ribeiro, Paula Gibrim, João Francisco Hecksher Olivetti, Daniel Mendes Barbosa, Julio C. S. Reis, and Philipe Melo. 2026. Só Mais Cinco Minutos: Uma Caracterização dos Jogadores Intensivos Brasileiros na Steam. In Anais do XXV Simpósio Brasileiro de Jogos e Entretenimento Digital (SBGames). SBC, Porto Alegre, RS, Brasil. In press.
Olga Russakovsky, Jia Deng, Hao Su, Jonathan Krause, Sanjeev Satheesh, Sean Ma, Zhiheng Huang, Andrej Karpathy, Aditya Khosla, Michael Bernstein, et al. 2015. ImageNet Large Scale Visual Recognition Challenge. International Journal of Computer Vision 115 (2015), 211–252.
Phillip Rust, Jonas F. Lotz, Emanuele Bugliarello, Elizabeth Salesky, Miryam de Lhoneux, and Desmond Elliott. 2023. PIXEL: Copy-Free Pixel-Based Language Modeling. In Advances in Neural Information Processing Systems (NeurIPS), Vol. 36. 43120–43135.
Lucinda Saldanha, Sofia Marques da Silva, and Pedro D Ferreira. 2023. “Community” in video game communities. Games and Culture 18, 8 (2023), 1004–1022. DOI: 10.1177/15554120221150058
Ranjan Sapkota, Rahul Harsha Cheppally, Ajay Sharda, and Manoj Karkee. 2025. YOLO26: key architectural enhancements and performance benchmarking for real-time object detection. arXiv preprint arXiv:2509.25164 (2025).
Rico Sennrich, Barry Haddow, and Alexandra Birch. 2016. Neural Machine Translation of Rare Words with Subword Units. In Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics. 1715–1725. [link]
Karen Simonyan and Andrew Zisserman. 2015. Very Deep Convolutional Networks for Large-Scale Image Recognition. arXiv:1409.1556 [cs.CV] [link]
Valve Corporation. 2026. Steam Web API. [link] Acesso em: 20 abr. 2026.
V Vijayakumar, PA Adolf, and D Hari Prasad. 2021. A novel approach for image based cyberbullying detection and prevention. Int J Comput Appl 183 (2021), 41–45.
Yonghui Wu, Mike Schuster, Zhifeng Chen, Quoc V Le, Mohammad Norouzi, Wolfgang Macherey, Maxim Krikun, Yuan Cao, Qin Gao, Klaus Macherey, Jeff Klingner, Apurva Shah, Melvin Johnson, Xiaobing Liu, Łukasz Kaiser, Stephan Gouws, Yoshikiyo Kato, Taku Kudo, Hideto Kazawa, Keith Stevens, George Kurian, Nishant Patil, Wei Wang, Cliff Young, Jason Smith, Jason Riesa, Alex Rudnick, Oriol Vinyals, Greg Corrado, Macduff Hughes, and Jeffrey Dean. 2016. Google’s neural machine translation system: Bridging the gap between human and machine translation. arXiv preprint arXiv:1609.08144 (2016).
Xuemiao Xu, Linling Zhang, and Tien-Tsin Wong. 2010. Structure-based ASCII art. In ACM SIGGRAPH 2010 papers. 1–10.
Sergey Berezin, Reza Farahbakhsh, and Noel Crespi. 2024. Read over the lines: Attacking llms and toxicity detection systems with ascii art to mask profanity. arXiv preprint arXiv:2409.18708 (2024).
Daniel Bicho, Artur Ferreira, and Nuno Datia. 2020. A Deep Learning Approach to Identify Not Suitable for Work Images. i-ETC: ISEL Academic Journal of Electronics, Telecommunications and Computers 6, 1 (2020), 1–9. DOI: 10.22266/ijies2022.0228.19
Falcon Dai and Zheng Cai. 2017. Glyph-aware Embedding of Chinese Characters. In Proceedings of the First Workshop on Subword and Character Level Models in NLP (SCLeM). Association for Computational Linguistics, 64–69.
Mustafa Eriş and Mustafa Kaya. 2023. Leveraging Deep Learning for Identification of Illicit Images in Digital Forensic Investigations. Traitement du Signal 40, 6 (December 2023), 2539–2552. DOI: 10.18280/ts.400617
Joseph L Fleiss. 1971. Measuring nominal scale agreement among many raters. Psychological bulletin 76, 5 (1971), 378.
Akira Fujisawa, Kazuyuki Matsumoto, Kazuki Ohta, Minoru Yoshida, and Kenji Kita. 2020. Ascii art classification model by transfer learning and data augmentation. In Fuzzy Systems and Data Mining VI: Proceedings of FSDM 2020. SAGE Publications 1 Oliver’s Yard, 55 City Road, London, EC1Y 1SP, 608–618.
Paula Gibrim, Vinicius Lodi, Gabriel Andrade, Marcus Ribeiro, Mauricio Gruppi, Daniel Barbosa, Philipe Melo, and Julio C. S. Reis. 2026. Toxic Steam: A Large-Scale Characterization and Prediction of Toxic Users in a Gaming Platform. In Proceedings of the AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment (AIIDE). In press.
Paula Gibrim, Marcus Vinicius Guerra Ribeiro, Lorena Moreira, Daniel Mendes Barbosa, Philipe Melo, and Julio C. S. Reis. 2026. O Fenômeno da Toxicidade em Jogos Digitais: Uma Caracterização da Comunidade Brasileira na Steam. In Anais do XXV Simpósio Brasileiro de Jogos e Entretenimento Digital (SBGames). SBC, Porto Alegre, RS, Brasil. In press.
Paul Hardy and Roman Czyborra. 2023. GNU Unifont (Version 15.0.04). [link]. Acessado em: 28 jun. 2026.
Devamardeep Hayatpur, Brian Hempel, Kathy Chen, William Duan, Philip Guo, and Haijun Xia. 2024. Taking ascii drawings seriously: How programmers diagram code. In Proceedings of the 2024 CHI Conference on Human Factors in Computing Systems. 1–16.
Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. 2015. Deep Residual Learning for Image Recognition. arXiv:1512.03385 [cs.CV] [link]
Sture Holm. 1979. A Simple Sequentially Rejective Multiple Test Procedure. Scandinavian Journal of Statistics 6, 2 (1979), 65–70.
Andrew Howard, Mark Sandler, Grace Chu, Liang-Chieh Chen, Bo Chen, Mingxing Tan, Weijun Wang, Yukun Zhu, Ruoming Pang, Vijay Vasudevan, Quoc V Le, and Hartwig Adam. 2019. Searching for MobileNetV3. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV). 1314–1324.
Shalini Jangra, Zaid Almahmoud, Suparna De, Gareth Tyson, Ehsan Ul Haq, and Nishanth Sastry. 2026. Understanding the Complexities of Responsibly Sharing NSFW Content Online. In Proceedings of the International AAAI Conference on Web and Social Media, Vol. 20. 1083–1096.
Qi Jia, Xiang Yue, Shanshan Huang, Ziheng Qin, Yizhu Liu, Bill Yuchen Lin, and Yang You. 2024. Visual perception in text strings. arXiv preprint arXiv:2410.01733 (2024).
Fengqing Jiang, Zhangchen Xu, Luyao Niu, Zhen Xiang, Bhaskar Ramasubramanian, Bo Li, and Radha Poovendran. 2024. Art-Prompt: ASCII Art-based Jailbreak Attacks against Aligned LLMs. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). Association for Computational Linguistics, Bangkok, Thailand, 15157–15173. DOI: 10.18653/v1/2024.acl-long.809
J. R. Landis and G. G. Koch. 1977. The measurement of observer agreement for categorical data. Biometrics 33, 1 (1977), 159–174.
Tsung-Yi Lin, Priya Goyal, Ross Girshick, Kaiming He, and Piotr Dollár. 2017. Focal loss for dense object detection. In Proceedings of the IEEE international conference on computer vision. 2980–2988.
Ilya Loshchilov and Frank Hutter. 2017. Decoupled weight decay regularization. arXiv preprint arXiv:1711.05101 (2017).
Kerry Luo, Michael Fu, Joshua Peguero, Husnain Malik, Anvay Patil, Joyce Lin, Megan Van Overborg, Ryan Sarmiento, and Kevin Zhu. 2025. ASCIIBench: Evaluating Language-Model-Based Understanding of Visually-Oriented Text. arXiv preprint arXiv:2512.04125 (2025).
Claude Nadeau and Yoshua Bengio. 2003. Inference for the Generalization Error. Machine Learning 52, 3 (2003), 239–281. DOI: 10.1023/A:1024068626366
Qwen Team. 2025. Qwen2.5 Technical Report. arXiv preprint arXiv:2412.15115 (2025).
Qwen Team. 2025. Qwen2.5-VL Technical Report. arXiv preprint arXiv:2502.13923 (2025).
Marcus Vinicius Guerra Ribeiro, Bernardo Cunha, Vitor Andrade, Henrique Campos, Daniel Mendes Barbosa, and Philipe Melo. 2026. Livre para Quem? Uma Análise das Discrepâncias entre a Classificação Etária Indicativa e o Conteúdo Adulto Autodeclarado na Steam. In Anais do XXV Simpósio Brasileiro de Jogos e Entretenimento Digital (SBGames). SBC, Porto Alegre, RS, Brasil. In press.
Marcus Vinicius Guerra Ribeiro, Paula Gibrim, João Francisco Hecksher Olivetti, Daniel Mendes Barbosa, Julio C. S. Reis, and Philipe Melo. 2026. Só Mais Cinco Minutos: Uma Caracterização dos Jogadores Intensivos Brasileiros na Steam. In Anais do XXV Simpósio Brasileiro de Jogos e Entretenimento Digital (SBGames). SBC, Porto Alegre, RS, Brasil. In press.
Olga Russakovsky, Jia Deng, Hao Su, Jonathan Krause, Sanjeev Satheesh, Sean Ma, Zhiheng Huang, Andrej Karpathy, Aditya Khosla, Michael Bernstein, et al. 2015. ImageNet Large Scale Visual Recognition Challenge. International Journal of Computer Vision 115 (2015), 211–252.
Phillip Rust, Jonas F. Lotz, Emanuele Bugliarello, Elizabeth Salesky, Miryam de Lhoneux, and Desmond Elliott. 2023. PIXEL: Copy-Free Pixel-Based Language Modeling. In Advances in Neural Information Processing Systems (NeurIPS), Vol. 36. 43120–43135.
Lucinda Saldanha, Sofia Marques da Silva, and Pedro D Ferreira. 2023. “Community” in video game communities. Games and Culture 18, 8 (2023), 1004–1022. DOI: 10.1177/15554120221150058
Ranjan Sapkota, Rahul Harsha Cheppally, Ajay Sharda, and Manoj Karkee. 2025. YOLO26: key architectural enhancements and performance benchmarking for real-time object detection. arXiv preprint arXiv:2509.25164 (2025).
Rico Sennrich, Barry Haddow, and Alexandra Birch. 2016. Neural Machine Translation of Rare Words with Subword Units. In Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics. 1715–1725. [link]
Karen Simonyan and Andrew Zisserman. 2015. Very Deep Convolutional Networks for Large-Scale Image Recognition. arXiv:1409.1556 [cs.CV] [link]
Valve Corporation. 2026. Steam Web API. [link] Acesso em: 20 abr. 2026.
V Vijayakumar, PA Adolf, and D Hari Prasad. 2021. A novel approach for image based cyberbullying detection and prevention. Int J Comput Appl 183 (2021), 41–45.
Yonghui Wu, Mike Schuster, Zhifeng Chen, Quoc V Le, Mohammad Norouzi, Wolfgang Macherey, Maxim Krikun, Yuan Cao, Qin Gao, Klaus Macherey, Jeff Klingner, Apurva Shah, Melvin Johnson, Xiaobing Liu, Łukasz Kaiser, Stephan Gouws, Yoshikiyo Kato, Taku Kudo, Hideto Kazawa, Keith Stevens, George Kurian, Nishant Patil, Wei Wang, Cliff Young, Jason Smith, Jason Riesa, Alex Rudnick, Oriol Vinyals, Greg Corrado, Macduff Hughes, and Jeffrey Dean. 2016. Google’s neural machine translation system: Bridging the gap between human and machine translation. arXiv preprint arXiv:1609.08144 (2016).
Xuemiao Xu, Linling Zhang, and Tien-Tsin Wong. 2010. Structure-based ASCII art. In ACM SIGGRAPH 2010 papers. 1–10.
Publicado
27/10/2026
Como Citar
LAGE, Kayo M.; FERNANDES, Thúlio M. O.; GIBRIM, Paula T. M.; RIBEIRO, Marcus V. G.; SILVA, Michel M.; MELO, Philipe de F.; REIS, Julio C. S..
Arte ou Abuso? Uma Abordagem Comparativa para Classificação de ASCII Art NSFW na Steam. In: BRAZILIAN SYMPOSIUM ON MULTIMEDIA AND THE WEB (WEBMEDIA), 32. , 2026, Lavras/MG.
Anais [...].
Porto Alegre: Sociedade Brasileira de Computação,
2026
.
p. 104-113.
DOI: https://doi.org/10.5753/webmedia.2026.30982.
