← Back to projects
kaggle · April 2024
NLP Disaster Tweets
Placeholder — distilled transformer for real-vs-fake disaster tweet classification.
Python Transformers DistilBERT NLP
outcome
F1 0.84
Problem
Placeholder — binary tweet classification under noisy labels.
Approach
Placeholder — DistilBERT fine-tuning with label-smoothing and hard-example mining.
Results
Placeholder — sustained F1 gains across cross-validation folds.