Validação de técnicas de data augmentation em corpus textuais sobre diabetes

Abstract

Diabetes is a chronic disease that affects millions of people worldwide. With the assistance of computational models, it is possible to aid in the control and even prevention of diabetes. These models can be based on machine learning methods that aim to recognize patterns of behavior, for example, in textual datasets from social networks, such as Twitter. However, the efficiency of these intelligent models is linked to the volume and quality of available data. In this context, the study proposes to validate data augmentation techniques in a corpus related to diabetes derived from Twitter, with the aim of establishing a reliable and high-quality source for future research in natural language processing applied to this specific area of health. The results demonstrate that in some samples of the corpus, the meaning of sentences was altered after data augmentation; however, this change was not significant enough to affect efficiency, ensuring that the augmented corpus performs better in text classification models.