Only the ABCD Team participated in the Aspect Extraction sub-task.  

They modeled the task as a token-classification problem with a BIO tagging scheme, fine-tuning several Transformer-based models.  

The best results were achieved using BERTimbau Large, a model trained exclusively for Portuguese.

Two teams participated in the Aspect Sentiment Classification sub-task: UFPR and ABCD.  

The best performance came from UFPR, who fine-tuned a BERT-based model on an augmented dataset generated with ChatGPT.  

The ABCD Team explored both generation and classification approaches, with the classification method performing best.