Classifying Fake News in Filipino Language using Improved DistilBERT Model

Detecting misinformation in the Filipino language poses unique challenges, especially when working with limited data resources. This study explored the use of transformer models, specifically pre-trained models like BERT, to construct a fake news classification model for Filipino. However, BERT's large size and computation hinder its deployment in resource-constrained environments. To address this, the study focuses on developing a lightweight fake news classifier using the DistilBERT architecture, fine-tuned for Filipinos. The model's performance was evaluated against BERT based on accuracy, precision, recall, F1 score, inference time, memory requirement, and parameter size. Results indicate that while BERT exhibits slightly better classification performance, DistilBERT offers significant advantages in efficiency with faster inference time and smaller resource requirements.