LLM-based augmentation raises Bangla fake news detection F1 score from 0.85 to 0.88 using synthetic data generation

Researchers used the instruction-tuned Gemma 3 27B IT model to generate 4,545 synthetic Bangla fake news samples, applying semantic filtering and controlled subsampling to maintain label consistency and diversity.

Augmenting only the minority class with high augmentation rate and random subsampling achieved the strongest performance gains, demonstrating that well-designed LLM-driven augmentation can improve fake news detection in low-resource languages.

The synthetic dataset and full implementation have been publicly released to support reproducibility and further research in multilingual misinformation detection.