Model description
This model is a version of digitalepidemiologylab/covid-twitter-bert-v2 fine-tuned on a monkeypox misinformation dataset available here.
Intended uses & limitations
The model is intended to be used for the binary classification of monkeypox related tweets into one of two classes: 'misinformation' or 'not misinformation'.
Training and evaluation data
The training dataset consists of 5,787 tweets with an imbalanced distribution of the two classes (in favour of the 'not misinformation' class). Training and evaluation data can be accessed here.
Training procedure
The model was trained for 20 epochs using an early-stopping callback and restoration of weights for the epoch with the lowest validation loss. The following hyperparameters were used during training:
- Initial learning rate: 5e-06
- Learning rate strategy: reduce on plateau (patience = 2)
- Batch size: 4
- Optimiser: Adam
Framework versions
- Transformers 4.21.1
- TensorFlow 2.8.2
- Datasets 2.4.0
- Tokenizers 0.12.1