Uso de LLMs no apoio à geração de strings de busca para o desenvolvimento de Estudos Secundários

• Ano de defesa: 2025
• Autor(a) principal: MARIA LUÍSA DE BARROS COSTA SILVA
• Orientador(a): Bruno Magalhaes Nogueira
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Fundação Universidade Federal de Mato Grosso do Sul
• País: Brasil
• Palavras-chave em Português: Estudos Secundários; Modelos de Linguagem; LLM
• Link de acesso: https://repositorio.ufms.br/handle/123456789/11654

Secondary Studies (SS) are a widely used methodology in the Software Engineering scientific field, since the introduction of Evidence-Based Software Engineering. The main objective of Secondary Studies is to gather all available information on a concept or phenomenon. One of the steps needed for the conduction of a SS is the definition and execution of a search strategy. One of the main strategies applied is the automated search, in order to perform this strategy, it is necessary to create and refine a search string that will be used in search engines. In the recent years, the textual technology domain has evolved greatly with the advance of Large language models (LLMs), which, through the transformers architecture and an expressive number of parameters, enable a high semantic performance combined with low complexity of use. Based on the difficulty in constructing search strings, this work proposes the creation of SeSG-LLM. SeSG-LLM is a tool based on the Alves et al. (2022) work, the Search String Generator (SeSG). The SeSGx-LLM version aims to integrate Large language models into the SeSG framework. In conclusion, the results demonstrated that LLMs can facilitate the generation of synonyms that will compose the strings, with Mistral 7B exhibiting the most consistent performance among the tested models. Additionally, the findings indicated that LDA demonstrated superior performance in the extraction of keywords.