| Resumen |
The following work aims to describe the team participation in JOKER 2024, which focuses on developing various methods for classifying text that exhibit different techniques and humorous intentions. Understanding such aspects of humor can often be challenging for human beings. By classifying humor into these categories, we aim to establish more robust methods for classification, which can be applied across various fields of study. Current models offer high potential for training and fine-tuning complex tasks like humor classification. This ranges from the traditional use of Convolutional Neural Networks (CNNs) to the widely utilized modern Transformer paradigm BERT-like models. The results were mixed, as different approaches were chosen. It is believed that, given their performance, the models can still be optimized and their accuracy improved. Overall, the results are satisfactory for a first approach using the usual BERT-like model and embeddings such a USE with a CNN. © 2024 Copyright for this paper by its authors. |
