Mineração de texto, inteligência artificial e aplicações em biotecnologia

• Ano de defesa: 2024
• Autor(a) principal: Eulálio Reis, Vítor
• Orientador(a): Caracelli, Ignez
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de São Carlos; Câmpus São Carlos
• Programa de Pós-Graduação: Programa de Pós-Graduação em Biotecnologia - PPGBiotec
• Palavras-chave em Português: Vitamina D; Covid-19; Análise de Proteínas (RNA_M); Inteligência Artificial
• Palavras-chave em Inglês: LLM; Sci-Bot; Vitamin D; Protein Analysis (RNA_M); Artificial Intelligence
• Área do conhecimento CNPq: CIENCIAS EXATAS E DA TERRA
• Link de acesso: https://hdl.handle.net/20.500.14289/21572

Large Language Models (LLMs) are AI-powered programs that generate and manipulate text, learning to understand and respond in a human-like way by developing their descriptive capabilities and potential. LLMs face challenges such as specific knowledge gaps, factuality issues, and hallucinations. Therefore, Retrieval-Augmented Generation (RAG) aims to address these challenges by connecting models to external and verified knowledge sources. In this context, the RAG approach is used here to analyze and extract responses from scientific texts, investigating the intrinsic relationship within the complex connections and the relationship with current use. The methods and techniques are based on AI for different language processing to retrieve, rationalize, and mobilize the knowledge present in an interdisciplinary technical-scientific knowledge to be validated. Therefore, techniques for text mining generated from relevant documents are integrated, while text generation seems to operate more effectively in complex contexts for specific in-depth analysis. Additionally, the field of Science or Science-Bot studies potential connections and interactions that are imagined or underrepresented in the literature, contributing to the ongoing debate in various ways. As a case study, Sci-Bot was fed scientific texts with existing information on possible directions for future research, public health strategies in the area control and prevention, and even in changing cultural traits. Thus, the RAG approach makes it possible to articulate textual information from the scientific literature, enabling a detailed explanation on an unprecedented scale of all the complex connections or intersections between the key concepts involved in the recent context of the pandemic.