Teaching Assistant Based on a Brazilian Large Language Model

• Elton S. Siqueira UFPA
• Carlos S. Portela UFPA
• Augusto N. Moraes UFPA

DOI: https://doi.org/10.5753/sbsi.2025.246480

Resumo

Context: The use of AI Teaching Assistants (AI-TAs) in educational contexts aims to support student learning in complex subjects, where there are challenges in assimilating interdisciplinary concepts and a need for quick and accurate responses. Problem: Students face barriers when studying subjects that require prior knowledge in multiple areas, leading to comprehension difficulties that negatively affect their academic progress. Proposed Solution: An AI-TA based on the Brazilian language model Sabiá 2.0 is proposed, incorporating mechanisms for Quick Verification, Expansion, and Error Correction to enhance the question-and-answer system. IS Theory: This work is grounded in Social Learning Theory, which explores how individuals acquire knowledge in technology-mediated interaction contexts. This theory is applicable to the use of AI-TAs, which act as learning mediators, promoting knowledge acquisition and facilitating understanding of complex concepts. Method: The methodology is Design Science Research, with a descriptive approach and quantitative analysis, including proof of concept and practical evaluation with users. The system was implemented in an educational environment and tested for performance and student satisfaction. Summary of Results: The system showed a high satisfaction rate regarding response quality; however, it revealed limitations in response time for model-based queries, suggesting improvements in API optimization. Contributions and Impact in IS Field: This study contributes to the IS field by demonstrating how AI models can enhance teaching and student support, proposing a robust and adaptable solution for learning environments.

Palavras-chave: Educational Support, TeachingAssistant, LLM, Question-Answering System

Referências

Imran Ahmed, Gwanggil Jeon e Francesco Piccialli. 2022. From artificial intelligence to explainable artificial intelligence in industry 4.0: a survey on what, how, and where. IEEE Transactions on Industrial Informatics, 18, 8, 5031–5042. DOI: 10.1109/TII.2022.3146552.

Thales Sales Almeida, Hugo Abonizio, Rodrigo Nogueira e Ramon Pires. 2024. Sabiá-2: a new generation of portuguese large language models. (2024). [link] arXiv: 2403.09887 [cs.CL].

Yiqian Chen, Hanxi Deng, Chi-Hua Chen e Chan-Liang Chung. 2023. Efficient artificial intelligence-teaching assistant based on chatgpt. Em 2023 International Conference on Smart Systems for applications in Electrical Sciences (ICSSES), 1–5. DOI: 10.1109/ICSSES58299.2023.10200077.

Yingjun Gong. 2021. Application of virtual reality teaching method and artificial intelligence technology in digital media art creation. Ecol. Informatics, 63, 101304. [link].

Yann Hicke, Anmol Agarwal, Qianou Ma e Paul Denny. 2023. Ai-ta: towards an intelligent question-answer teaching assistant using open-source llms. (2023). [link] arXiv: 2311.02775 [cs.LG].

Stefan Huber, Kristian Kiili, Steve Nebel, Richard Ryan, Michael Sailer e Manuel Ninaus. 2024. Leveraging the potential of large language models in education through playful and game-based learning. Educational Psychology Review, 36, (fev. de 2024). DOI: 10.1007/s10648-024-09868-z.

Jaeho Jeon e Seongyong Lee. 2023. Large language models in education: a focus on the complementary relationship between human teachers and chatgpt. Education and Information Technologies, (mai. de 2023), 0000. DOI: 10.1007/s10639-023-11834-1.

Jinhee Kim, Hyunkyung Lee e Young Cho. 2022. Learning design to support student-ai collaboration: perspectives of leading teachers for ai in education. Education and Information Technologies, 27, (jan. de 2022), 1–36. DOI: 10.1007/s10639-021-10831-6.

Treice Moreira, Cláudia Silva, Cláudio Passos, Isabel Fernandes e Ronaldo Goldschmidt. 2023. Tutor inteligente em jogo educacional digital para capacitação na identificação de fake news em português: experimentos preliminares. Em Anais Estendidos do XXXIV Simpósio Brasileiro de Informática na Educação. SBC, Passo Fundo/RS, 14–20. DOI: 10.5753/sbie_estendido.2023.235269.

Ramon Pires, Hugo Abonizio, Thales Sales Almeida e Rodrigo Nogueira. 2023. [inline-graphic not available: see fulltext] sabiá: portuguese large language models. Em Intelligent Systems: 12th Brazilian Conference, BRACIS 2023, Belo Horizonte, Brazil, September 25–29, 2023, Proceedings, Part III. Springer-Verlag, Belo Horizonte, Brazil, 226–240. isbn: 978-3-031-45391-5. DOI: 10.1007/978-3-031-45392-2_15.

Evandro Preuss, Dante Barone e Renato Henriques. 2020. Uso de técnicas de inteligência artificial num sistema de mesa tangível. Em Anais do XXVI Workshop de Informática na Escola. SBC, Evento Online, 439–448. DOI: 10.5753/cbie.wie.2020.439.

James P. Smith. 2015. A case study on design science research as a methodology for developing tools to support lean construction efforts. English. Em 23rd Annual Conference of the International Group for Lean Construction. Olli Seppänen, Vicente A. González e Paz Arroyo, editores. Perth, Australia, 517–526. [link].

Hanyi Xu, Wensheng Gan, Zhenlian Qi, Jiayang Wu e Philip Yu. 2024. Large language models for education: a survey. (Mai. de 2024). DOI: 10.48550/arXiv.2405.13001.

Fernando Zaidan, Marcello Bax e Fernando Silva Parreiras. 2016. Design science research: aplicação em um projeto de pesquisa e desenvolvimento. Em 13th International Conference on Information Systems and Technology Management - CONTECSI. (Jun. de 2016), 3757–3774. DOI: 10.5748/9788599693124-13CONTECSI/PS-4163.