Fine-tuned google-bert-uncased model
The model is fine-tuned using 210k huffington post articles into 42 categories.
The model can handle a maximum token length of 256
The model was fine-tuned using 60 / 20 / 20 stratified splits for training, eval and test sets
The model achieves a performance of:
eval_loss: 1.968
eval_f1: 0.786
epoch: 10.0
The model performance based on weighted F1-score across 42 categories is:
Weighted F1 Score on Test Set: 0.7126
Examples Predictions:
Text: Manage Your Expectations, Lower Your Stress
Your outlook is a product of your own relationship with expectation. What will or won't happen, no one knows. And how we deal with the stress of not knowing, whether to hope for the best or expect the worst, the idea that our expectations always directly affect an outcome is little more than magical thinking.
Predicted label: WELLNESS, Score: 0.8339
Text: Donald Trump Tells Drought-Plagued Californians: 'There Is No Drought'
"If I win, believe me, we’re going to start opening up the water."
Predicted label: POLITICS, Score: 0.8151
Text: Sticker Shock: 5 Of The World's Most Expensive Hotel Suites (PHOTOS)
You can only imagine our heart palpitations when we see the nightly rates of these hotels' top-of-the-line suites.
Predicted label: TRAVEL, Score: 0.9901
- Downloads last month
- -