metadata
language: en
tags:
- go-emotion
- text-classification
- pytorch
datasets:
- go_emotions
metrics:
- f1
widget:
- text: Thanks for giving advice to the people who need it! 👌🙏
license: mit
Model Description
- Based on the uncased BERT pretrained model with a linear output layer.
- Added several commonly-used emoji and tokens to the special token list of the tokenizer.
- Did label smoothing while training.
- Used weighted loss and focal loss to help the cases which trained badly.
Results
Best Result of Macro F1 - 53%