|
--- |
|
language: |
|
- en |
|
license: lgpl-3.0 |
|
tags: |
|
- text-classification |
|
- transformers |
|
- tf |
|
metrics: |
|
- accuracy |
|
- f1 |
|
- precision |
|
- recall |
|
--- |
|
|
|
# distilbert-political-tweets |
|
|
|
🗣 🇺🇸 Classify sentiment as Democratic or Republican. |
|
|
|
# Model |
|
|
|
[DistilBERT base model (uncased)](https://huggingface.co/distilbert-base-uncased) for sequence classification fine-tuned on ~10,000 randomly selected tweets from Democratic and Republican senators from FiveThirtyEight's [twitter-ratio/senators](https://fivethirtyeight.datasettes.com/fivethirtyeight/twitter-ratio%2Fsenators#export) dataset. |
|
|
|
# Evaluation |
|
|
|
The model was evaluated on a validation dataset of ~3,000 unseen random tweets: |
|
```python |
|
{'eval_accuracy': 0.9348, |
|
'eval_f1': 0.9386, |
|
'eval_loss': 0.1975, |
|
'eval_precision': 0.9281, |
|
'eval_recall': 0.9494} |
|
``` |
