Base de conhecimento para recuperação e acesso à informação acadêmica: provimento para um chatbot na ECI UFMG

Abstract

The retrieval of academic information (regulations, resolutions and procedures) in public universities poses a challenge for students, given the dispersion of documents across multiple channels and their technical language. In this context, informational curation emerges as a strategy for organization, while the knowledge base provides a structured means to support retrieval in systems, these being the core questions guiding this research. In light of this gap, the present dissertation aimed to structure and implement a knowledge base to support the development of a chatbot designed for the retrieval of academic information (regulations, resolutions and procedures) about undergraduate at the School of Information Science of the Federal University of Minas Gerais. Specifically, it sought to: (i) map methods and techniques used in the development of information-oriented chatbots; (ii) identify the information-seeking behavior of the students regarding normative documents; (iii) identify and analyze implemented educational chatbots; and (iv) conduct informational curation to support the implementation of a knowledge base. This research is characterized as a case study, with an applied, descriptive and exploratory nature, adopting a mixed-methods approach, structured in four stages. The first stage consisted of a literature review on chatbot development, identifying methodologies, architectures, languages, platforms and Natural Language Processing techniques. It was observed that existing studies privilege technical aspects, often neglecting information treatment. The second stage involved the analysis of a questionnaire administered within the project Information Retrieval in Artificial Intelligence Models, in 2025, with responses from undergraduate students at UFMG, which allowed for an understanding of how students seek normative documents, the challenges that they face, as well as their doubts and opinions. The results indicated that the fragmentation of information sources and the complexity of normative language constitute the main difficulties. In the third stage, studies on educational chatbots, previously identified in the literature review, were analyzed and selected based on the presentation of documented and replicable models. The analysis provided examples of interaction flows, intent structures, uses of entities and knowledge bases, serving as practical references. Finally, the fourth stage focused on the informational curation of the knowledge base, developed in five phases: identification of informational needs; selection of sources and documents; structuring and creation of the base; development (population and validation); and modeling of informational flows. This stage resulted in a knowledge base comprising 31 core concepts, organized into a taxonomic structure with 225 intents (questions and answers) grounded in official documents, each accompanied by exact references to articles, paragraphs and document links. Although validation involved an expert, the absence of end-user validation was a limitation of the study and remains as future work. This research made a practical contribution by presenting a case study that generated a product: a structured and scalable knowledge base that offers theoretical and methodological grounding, within the informational context, for the development of innovative solutions, fostering digital transformation and promoting social impact. This strategic product will support the implementation of an institutional chatbot capable of providing contextualized, reliable and user-centered responses aligned with the informational needs of the students.