Large-scale characterization of financial scams and fraud based on bank complaints from Reclame Aqui

• Thalia B. M. de Almeida UFMG
• Karen Gomes UFMG
• Fabrício Benevenuto UFMG
• Marcos André Gonçalves UFMG
• Antonio A. F. Loureiro UFMG

DOI: https://doi.org/10.5753/webmedia.2026.30847

Abstract

Financial fraud has become a critical challenge in Brazil’s digital banking ecosystem, yet large-scale evidence on how victims describe scams and unauthorized fraud in Portuguese remains scarce. This paper presents the first large-scale characterization of banking fraud narratives from Reclame Aqui, investigating recurring fraud mechanisms, social engineering strategies, and institutional responses. To enable this analysis, we develop a hybrid pipeline combining weak supervision, LLM-based annotation, and supervised fine-tuning of BERTimbau. Applied to a corpus of 221,519 complaints, the pipeline achieves a Macro F1 of 0.849 and identifies 20,127 candidate reports of scams and unauthorized fraud. BERTopic reveals 19 recurring fraud mechanisms, split between social engineering scams and unauthorized fraud. Pix-based social engineering emerges as the dominant pattern (33.9%), bringing together diverse impersonation schemes that converge on immediate, irreversible transfers. Further analyses show that trust and greed are the predominant psychological levers, primarily associated with phone calls and social media, respectively, while scams exhibit substantially lower resolution rates than unauthorized fraud (16.7% vs. 29.3%). Overall, our findings demonstrate that consumer complaint platforms constitute a valuable and underexplored resource for large-scale financial fraud characterization, providing empirical evidence on fraud mechanisms, social engineering strategies, and institutional responses that can support fraud prevention, regulatory analysis, and future research on financial crime in Brazilian Portuguese.

Keywords: golpes, fraude, processamento de linguagem natural, BERTimbau, anotação por LLM, Reclame Aqui, modelagem de tópicos

References

Banco Central do Brasil. 2025. IF.data: Dados Selecionados de Entidades Supervisionadas. [link]. Data-base: 12/2025. Acesso em: 19/08/2026.

Kaveh Bastani, Hamed Namavari, and Jeffry Shaffer. 2019. Latent Dirichlet Allocation (LDA) for Topic Modeling of the CFPB Consumer Complaints. Expert Systems with Applications 127 (2019), 256–271. arXiv:1807.07468 [cs.IR] DOI: 10.1016/j.eswa.2019.03.001

Petros Boulieris, John Pavlopoulos, et al. 2023. Fraud Detection with Natural Language Processing. Machine Learning (2023). DOI: 10.1007/s10994-023-06354-5

Gerlof Bouma. 2009. Normalized (Pointwise) Mutual Information in Collocation Extraction. In Proceedings of the Biennial GSCL Conference. 31–40.

Jo Braithwaite. 2024. “Authorized Push Payment” Bank Fraud: What Does an Effective Regulatory Response Look Like? Journal of Financial Regulation 10, 2 (2024), 174–193. DOI: 10.1093/jfr/fjae006

Brasil. 1940. Decreto-Lei n. 2.848, de 7 de dezembro de 1940. Código Penal. [link] Acesso em: 28 jun. 2026.

Mark Button, Chris Lewis, and Jacki Tapley. 2009. Fraud Typologies and the Victims of Fraud: Literature Review. Technical Report. National Fraud Authority, London. [link] Acesso em: 28 jun. 2026.

Robert B. Cialdini. 2006. Influence: The Psychology of Persuasion (revised ed.). Harper Business.

Washington Cunha, Celso França, Guilherme Fonseca, Leonardo Rocha, and Marcos André Gonçalves. 2023. An Effective, Efficient, and Scalable Confidence-Based Instance Selection Framework for Transformer-Based Text Classification. In Proceedings of the 46th International ACM SIGIR Conference on Research and Development in Information Retrieval. ACM, 665–674. DOI: 10.1145/3539618.3591638

Washington Cunha, Felipe Viegas, Celso França, Guilherme Fonseca, Leonardo Rocha, and Marcos André Gonçalves. 2025. A Noise-Oriented and Redundancy-Aware Instance Selection Framework. ACM Transactions on Information Systems 43, 1 (2025). DOI: 10.1145/3701769

Dados Bancários. 2026. Ranking de Instituições por Ativo Total – Maiores Bancos do Brasil. [link] Dados compilados a partir do sistema IF.data do Banco Central do Brasil.

European Union Agency for Cybersecurity. s.d.. Threat Landscape. [link]. Acesso em: 29 jun. 2026.

Febraban. 2024. Comitê Executivo da Febraban lança Selo de Prevenção a Fraudes das Instituições Financeiras. Federação Brasileira de Bancos. [link] Inclui estimativa de que cerca de 70% dos golpes são de engenharia social.

Febraban. 2025. Golpe da falsa venda é o mais aplicado contra clientes bancários no 1o semestre do ano. Federação Brasileira de Bancos. [link]

David Fernandes, Edleno Silva de Moura, Berthier A. Ribeiro-Neto, Altigran Soares da Silva, and Marcos André Gonçalves. 2007. Computing Block Importance for Searching on Web Sites. In Proceedings of the Sixteenth ACM Conference on Information and Knowledge Management. ACM, 165–174. DOI: 10.1145/1321440.1321466

Maria Beatriz Giusti. 2025. Mais de 24 milhões de pessoas foram vítimas de golpes pelo Pix em um ano. Senado Notícias. [link] Acesso em: 28 jun. 2026.

Maarten Grootendorst. 2022. BERTopic: Neural Topic Modeling with a Class-based TF-IDF Procedure. arXiv preprint arXiv:2203.05794 (2022).

J. Richard Landis and Gary G. Koch. 1977. The Measurement of Observer Agreement for Categorical Data. Biometrics 33, 1 (1977), 159–174. DOI: 10.2307/2529310

Terri Miller. 2023. Slow your scroll: Spot and avoid social media giveaway scams. [link]. Federal Trade Commission. Acesso em: 29 jun. 2026.

National Institute of Standards and Technology. s.d.. Phishing. [link]. NIST Computer Security Resource Center Glossary. Acesso em: 29 jun. 2026.

Antonis Papasavva, Max Aliapoulios, et al. 2024. Applications of AI-Based Models for Online Fraud Detection and Analysis. Crime Science (2024). Systematic Literature Review.

Gabriel Prenassi, Ana Machado, Ester Souza, Mateus Brito, Davi Reis, Jessica Costa, Geovana Oliveira, Carlos Ferreira, and Leonardo Rocha. 2025. Narrativas de Jogos de Azar em Plataformas de Vídeo: Um Estudo Linguístico-Temático sobre o jogo do Tigrinho no YouTube. In Anais do XXXI Simpósio Brasileiro de Sistemas Multimídia e Web. Sociedade Brasileira de Computação, Porto Alegre, RS, Brasil, 366–375. DOI: 10.5753/webmedia.2025.15565

Alexander Ratner, Stephen H. Bach, Henry Ehrenberg, Jason Fries, Sen Wu, and Christopher Ré. 2017. Snorkel: Rapid Training Data Creation with Weak Supervision. Proceedings of the VLDB Endowment 11, 3 (2017), 269–282. DOI: 10.14778/3157794.3157797

Reclame AQUI. 2026. Como funciona. [link] Acesso em: 28 jun. 2026.

Michael Röder, Andreas Both, and Alexander Hinneburg. 2015. Exploring the Space of Topic Coherence Measures. In Proceedings of WSDM. 399–408.

Said Salloum, Tarek Gaber, Sunil Vadera, and Khaled Shaalan. 2022. A Systematic Literature Review on Phishing Email Detection Using Natural Language Processing Techniques. IEEE Access 10 (2022), 65703–65727. DOI: 10.1109/ACCESS.2022.3183083

Vasudeva Raju Sangaraju, Bharath Kumar Bolla, Deepak Kumar Nayak, and Jyothsna Kh. 2022. Topic Modelling on Consumer Financial Protection Bureau Data: An Approach Using BERT Based Embeddings. arXiv:2205.07259 [cs.LG] DOI: 10.48550/arXiv.2205.07259

Serasa Experian. 2025. Mais da metade dos brasileiros já foi vítima de fraude e 20% deles perderam até R$ 5 mil, revela estudo inédito da Serasa Experian. Serasa Experian. [link]

Serasa Experian. 2025. Relatório de Identidade Digital e Fraude 2025. [link]

Frank Stajano and Paul Wilson. 2011. Understanding Scam Victims: Seven Principles for Systems Security. Commun. ACM 54, 3 (2011), 70–75. DOI: 10.1145/1897852.1897872

Superior Tribunal de Justiça. 2012. Súmula 479. [link] As instituições financeiras respondem objetivamente pelos danos gerados por fortuito interno relativo a fraudes e delitos praticados por terceiros no âmbito de operações bancárias. Acesso em: 28 jun. 2026.

Superior Tribunal de Justiça. 2025. Bancos e instituições de pagamento devem indenizar clientes por falhas que viabilizam golpe da falsa central. [link] Notícia institucional referente aos REsp 2.222.059 e 2.229.519. Acesso em: 28 jun. 2026.

Zeya Lwin Tun and Daniel Birks. 2023. Supporting Crime Script Analyses of Scams with Natural Language Processing. Crime Science 12, 1 (2023). DOI: 10.1186/s40163-023-00177-4

Felipe Viegas, Sérgio Canuto, Christian Gomes, Washington Cunha, Thierson Rosa, Sabir Ribas, Leonardo Rocha, and Marcos André Gonçalves. 2019. CluWords: Exploiting Semantic Word Clustering Representation for Enhanced Topic Modeling. In Proceedings of the Twelfth ACM International Conference on Web Search and Data Mining. ACM, 753–761. DOI: 10.1145/3289600.3291032

Felipe Viegas, Washington Cunha, Christian Gomes, Antônio Pereira, Leonardo Rocha, and Marcos André Gonçalves. 2020. CluHTM: Semantic Hierarchical Topic Modeling Based on CluWords. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics. Association for Computational Linguistics, 8138–8150. DOI: 10.18653/v1/2020.aclmain.724

Felipe Viegas, Washington Cunha, Christian Gomes, Leonardo Rocha, and Marcos André Gonçalves. 2025. Exploiting Contextual Embeddings in Hierarchical Topic Modeling and Investigating the Limits of the Current Evaluation Metrics. Computational Linguistics (2025).

Xuezhi Wang, Jason Wei, Dale Schuurmans, Quoc Le, Ed Chi, Sharan Narang, Aakanksha Chowdhery, and Denny Zhou. 2023. Self-Consistency Improves Chain of Thought Reasoning in Language Models. In Proceedings of ICLR.

Caleb Ziems, William Held, Omar Shaikh, Jiaao Chen, Zhehao Zhang, and Diyi Yang. 2024. Can Large Language Models Transform Computational Social Science? Computational Linguistics 50, 1 (2024), 237–291. DOI: 10.1162/coli_a_00502