Arthur Buzelin
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Arthur Buzelin has a total of 25 co-authors. Here are the top co-authors by number of shared publications:
- Wagner Meira
Universidade Federal de Minas Gerais, Belo Horizonte, Minas Gerais, Brazil
6 shared publications - Yan Aquino
Universidade Federal de Minas Gerais (UFMG), Belo Horizonte, Brazil
6 shared publications - Gisele Lobo Pappa
Universidade Federal de Minas Gerais, Belo Horizonte, Brazil
5 shared publications - Pedro Bento
Universidade Federal de Minas Gerais (UFMG), Belo Horizonte, Brazil
5 shared publications - Samira Malaquias
Universidade Federal de Minas Gerais (UFMG), Belo Horizonte, Brazil
4 shared publications - Virgilio Almeida
UFMG
4 shared publications
Latest Publications
- Open Access
- 5 September 2026
- Published by Association for Computing Machinery (ACM)
- p. 101-107
YouTube Discourse During the Iran Conflict and the Afghanistan Withdrawal
War discourse on YouTube emerges through linked interactions among videos, channels, audiences, comments, and replies. This paper compares YouTube discussions of two U.S.-linked conflicts: the Afghanistan withdrawal in 2021 and the Iran conflict escalation in 2026. We analyze four corpora separating news organizations from political influencers, comprising 340,383 comments from 212,968 unique commenters. Our framework combines channel-level discourse embeddings adapted from WEAT-style semantic association tests, and classifier-based toxicity and hate-related proxy scores. The results show that Afghanistan and Iran discussions form distinct discourse spaces, while some ideologically different actors converge toward similar patterns of conflict commentary. Semantic associations around geopolitical terms vary across conflicts and source types, indicating that the same vocabulary is framed differently across communities. Hostility analysis further shows that influencer-centered spaces tend to attract more toxic discussion, while the Iran corpus exhibits stronger and more persistent hate-related signals, including in replies.
- 5 September 2026
- Published by Association for Computing Machinery (ACM)
- p. 101-107
Open AccessYouTube Discourse During the Iran Conflict and the Afghanistan Withdrawal
War discourse on YouTube emerges through linked interactions among videos, channels, audiences, comments, and replies. This paper compares YouTube discussions of two U.S.-linked conflicts: the Afghanistan withdrawal in 2021 and the Iran conflict escalation in 2026. We analyze four corpora separating news organizations from political influencers, comprising 340,383 comments from 212,968 unique commenters. Our framework combines channel-level discourse embeddings adapted from WEAT-style semantic association tests, and classifier-based toxicity and hate-related proxy scores. The results show that Afghanistan and Iran discussions form distinct discourse spaces, while some ideologically different actors converge toward similar patterns of conflict commentary. Semantic associations around geopolitical terms vary across conflicts and source types, indicating that the same vocabulary is framed differently across communities. Hostility analysis further shows that influencer-centered spaces tend to attract more toxic discussion, while the Iran corpus exhibits stronger and more persistent hate-related signals, including in replies.
- Open Access
- 23 March 2026
- Published by Association for Computing Machinery (ACM)
- p. 1899-1906
From Classrooms to Feeds: Youth Activity on Bluesky Amid Brazil's School Smartphone Ban
- 23 March 2026
- Published by Association for Computing Machinery (ACM)
- p. 1899-1906
Open AccessFrom Classrooms to Feeds: Youth Activity on Bluesky Amid Brazil's School Smartphone Ban
- Open Access
- 23 March 2026
- Published by Association for Computing Machinery (ACM)
- p. 1891-1898
Framing and Bias in Shared Media: Cross-Platform Analysis of Polarized Political Communities
- 23 March 2026
- Published by Association for Computing Machinery (ACM)
- p. 1891-1898
Open AccessFraming and Bias in Shared Media: Cross-Platform Analysis of Polarized Political Communities
- Open Access
- 11 July 2025
- Published by Sociedade Brasileira de Computação in Revista Eletrônica de Iniciação Científica em Computação
- Vol. 23, 118-123
A Machine Learning-Guided Approach for a Multi-Epitope HIV Vaccine Design
Developing an effective HIV vaccine remains challenging due to the virus's variability and complex immune responses. We propose a novel multi-epitope vaccine design using machine learning and computational methods to identify conserved, immunodominant epitopes from diverse HIV variants. These epitopes, selected to elicit humoral and cellular responses -- targeting CD4+ T cells, CD8+ cytotoxic T cells, and B cells -- are incorporated into a chimeric protein delivered via a viral vector to enhance immunity. Our framework integrates epitope selection, in silico physicochemical predictions, 3D construction of the chimeric protein, illustrated in Figure 1, and in vitro analysis, contributing to the development of a broadly protective and durable HIV vaccine.
- 11 July 2025
- Published by Sociedade Brasileira de Computação in Revista Eletrônica de Iniciação Científica em Computação
- Vol. 23, 118-123
Open AccessA Machine Learning-Guided Approach for a Multi-Epitope HIV Vaccine Design
Developing an effective HIV vaccine remains challenging due to the virus's variability and complex immune responses. We propose a novel multi-epitope vaccine design using machine learning and computational methods to identify conserved, immunodominant epitopes from diverse HIV variants. These epitopes, selected to elicit humoral and cellular responses -- targeting CD4+ T cells, CD8+ cytotoxic T cells, and B cells -- are incorporated into a chimeric protein delivered via a viral vector to enhance immunity. Our framework integrates epitope selection, in silico physicochemical predictions, 3D construction of the chimeric protein, illustrated in Figure 1, and in vitro analysis, contributing to the development of a broadly protective and durable HIV vaccine.
- Open Access
- 19 May 2025
- Published by Association for Computing Machinery (ACM)
- p. 272-282
Analyzing Political Discourse on Discord during the 2024 U.S. Presidential Election
- 19 May 2025
- Published by Association for Computing Machinery (ACM)
- p. 272-282
Open AccessAnalyzing Political Discourse on Discord during the 2024 U.S. Presidential Election
- Open Access
- 17 November 2024
- Published by Sociedade Brasileira de Computação
- p. 23-31
A Change in Perspective: The Trade-Off Between Perspective API and Custom Models in Classifying Hate Speech in Portuguese
This paper examines the performance of the Perspective API, developed by Jigsaw, in detecting hate speech in Portuguese. Although the Perspective API supports multiple languages, its performance metrics are often aggregated, obscuring specific details. Our study reveals that the API’s AUC-ROC score for Portuguese is significantly lower than for English (0.744 vs. 0.942). To address this, we developed a BERT classifier model trained on a Portuguese Twitter hate speech dataset. Our model, with just 100 messages in it’s training set, outperformed the Perspective API. These findings highlight the need for more granular performance metrics and suggest that custom models may offer better solutions for specific languages.
- 17 November 2024
- Published by Sociedade Brasileira de Computação
- p. 23-31
Open AccessA Change in Perspective: The Trade-Off Between Perspective API and Custom Models in Classifying Hate Speech in Portuguese
This paper examines the performance of the Perspective API, developed by Jigsaw, in detecting hate speech in Portuguese. Although the Perspective API supports multiple languages, its performance metrics are often aggregated, obscuring specific details. Our study reveals that the API’s AUC-ROC score for Portuguese is significantly lower than for English (0.744 vs. 0.942). To address this, we developed a BERT classifier model trained on a Portuguese Twitter hate speech dataset. Our model, with just 100 messages in it’s training set, outperformed the Perspective API. These findings highlight the need for more granular performance metrics and suggest that custom models may offer better solutions for specific languages.