Uso da LLM como ferramenta para auxiliar no processo de Revisão Sistemática

Abstract

A systematic review consists of a scientific investigation into a research topic. One of the biggest challenges when starting a systematic review is defining the protocol, especially choosing the search terms and selecting relevant articles in the first rounds of research. This is because the topic is possibly still new to the researcher and their knowledge of the subject has yet to be explored in greater depth. In this context, it is believed that the use of LLMs (Large language models), which consist of NLP (Natural Language Processing) models trained on a large scale of textual data, can help researchers at this stage of a Systematic Review (or any other literature review). Against this scenario, this work presents a tool developed using LLM in the form of a chatbot that plays an initial role in defining/finding relevant synonyms for the search terms, as well as identifying relevant articles based on the search criteria. Currently, the search is only carried out in the Pubmed database and for the synonyms and articles returned, the chatbot offers the option of explaining why that article is considered relevant. An initial comparison of the relevant terms and articles returned with an already defined protocol showed that the tool is indeed capable of helping the researcher, having led a layperson to a search term similar to the one defined in the protocol in a few minutes.