distilbert-political-tweets 🗣 🇺🇸

This model is a fine-tuned version of distilbert-base-uncased on the m-newhauser/senator-tweets dataset, which contains all tweets made by United States senators during the first year of the Biden Administration. It achieves the following results on the evaluation set:

• Accuracy: 0.9076
• F1: 0.9117

Model description

The goal of this model is to classify short pieces of text as having either Democratic or Republican sentiment. The model was fine-tuned on 99,693 tweets (51.6% Democrat, 48.4% Republican) made by US senators in 2021.

Model accuracy may not hold up on pieces of text longer than a tweet.

Training hyperparameters

The following hyperparameters were used during training:

• optimizer: Adam
• training_precision: float32
• learning_rate = 5e-5
• num_epochs = 5

Framework versions

• Transformers 4.16.2
• TensorFlow 2.8.0
• Datasets 1.18.3
• Tokenizers 0.11.6