Regresar

Spanish Pre-Trained CaTrBETO Model for Sentiment Classification in Twitter

Abstract:

The classification and analysis of sentiments allow us to study the behavior of human beings concerning a specific topic. Our work proposes using a multimodal trained model for sentiment classification in Spanish, incorporating tweets. Our model is called CaTrBETO, an approach based on a caption transformer (CaTr) and BETO (BERT model trained on an extensive Spanish corpus). The proposed model allows us to understand the sentiments in tweets through the combination of images and text. We test our approach using the Ecuadorian prison crisis 2021-2022, obtaining good results without extensive data. Furthermore, we demonstrate that our model exceeds the accuracy of BETO's previous work.