sentiment-roberta-base

Fine-tuned RoBERTa-base for binary sentiment classification on the Sentiment140 dataset (1.6M tweets).

Base model

FacebookAI/roberta-base — the original RoBERTa-base from Liu et al. (2019), 125M parameters.

Training

• Dataset: Sentiment140 (1.6M tweets, 80/20 split, seed 42)
• Hyperparameters: learning rate 2e-5, batch size 16, 3 epochs
• Hardware: NVIDIA A10G, AWS SageMaker (g5.2xlarge)
• Training time: 7.5 hours
• Trainer: Hugging Face Transformers + Trainer API; load_best_model_at_end=True

Test set performance

Metric	Value
Accuracy	89.11%
Precision	0.901
Recall	0.879
F1	0.890

Intended use

Demonstration model for an academic purposes

Limitations

• English only, binary sentiment, 2009-era Twitter language.
• Sentiment140 labels generated automatically using emoticons (distant supervision), introducing systematic noise.
• Does not handle sarcasm reliably (the dataset does not separate it as a phenomenon).