Redes convolucionais em grafos aplicado à estimativa de esforço de software por analogia a partir de requisitos textuais

Abstract

This work explores the application of advanced machine learning and natural language processing methods for software effort estimation by analogy in software projects. Effort estimation is a crucial task that directly impacts project planning, deadlines, and budgets. Traditionally, methods based on analogy or expert judgment have been used, however, these methods often lack accuracy due to the subjectivity and complexity involved. This study is justified by the need to improve the accuracy of these estimates through the use of deep learning models. The main objective is to implement, train, and evaluate a BertGCN model, which combines the capabilities of the BERT language model with Graph Convolutional Networks (GCNs). The methodology includes preparing a dataset of software requirements, obtaining embeddings with BERT, constructing the BertGCN model, and comparing its performance with the SE3M model using metrics such as Mean Absolute Error (MAE), Mean Squared Error (MSE), and Median Absolute Error (MdAE). The conclusion highlights the importance of the research and the use of deep learning models, such as BertGCN, in software effort estimation.