Inteligência Artificial aplicada ao processamento de extratos de contratos de inovação aberta: uma comparação de APIS

Abstract

This study evaluates the application of artificial intelligence APIs in processing and classifying extracts from open innovation contracts, a crucial area for collaboration between academia and industry. The analyzed APIs include ChatGPT (OpenAI), LLaMA (Meta), and Mistral (Mistral AI), using metrics such as accuracy, precision, recall, and F1-Score. The methodology involved the practical implementation of the APIs, the creation of a benchmark, and a quantitative performance comparison. The results highlight the strengths and limitations of each tool, contributing to the development of automated solutions for managing contract extracts in this context.The findings indicate that for scenarios requiring high Precision (avoiding false positives), LLaMA may be the best choice. For scenarios that demand greater overall balance, ChatGPT stands out. Mistral may be useful in applications prioritizing consistency and computational efficiency. These results provide an overview of the suitability of different language models for the specific needs of this domain, guiding API selection based on the desired scenario. Furthermore, the findings can assist in developing more effective methodologies and the practical implementation of these technologies on platforms such as ipdelve.com.