SentimentAnalysis-bert-base-uncased-finetuned-emotion

This model is a fine-tuned version of bert-base-uncased for emotion classification of short English texts such as tweets and social media posts.

The model predicts one of six emotion classes:

• sadness
• joy
• love
• anger
• fear
• surprise

The model was trained using the 🤗 Transformers Trainer API.

---

Model performance

Evaluation results on the test set:

• Loss: 0.2913
• F1 Macro: 0.8801
• Accuracy: 0.9215

Macro F1 is reported as the primary metric because the dataset is imbalanced and this metric better reflects performance across all emotion classes.

---

Intended uses

This model is suitable for:

• Emotion classification of tweets and short social media texts
• NLP research and academic projects
• Emotion-aware chatbots
• Sentiment and emotion analytics dashboards

---

Limitations

• Optimized for short texts; performance may degrade on long documents
• English-only
• May reflect biases present in the training data
• Not intended for high-stakes or sensitive decision-making

---

Training data

The model was trained on an emotion-labeled dataset of short texts with six emotion categories.

Preprocessing steps included:

• Train / validation / test split
• Tokenization using the BERT tokenizer
• Padding and truncation to a fixed maximum sequence length
• Label encoding using Hugging Face ClassLabel

---

Training procedure

Hyperparameters

• Base model: bert-base-uncased
• Learning rate: 1e-5
• Train batch size: 16
• Eval batch size: 16
• Epochs: 6
• Optimizer: AdamW (Torch fused)
  • betas = (0.9, 0.999)
  • epsilon = 1e-8
• Learning rate scheduler: Linear
• Seed: 42

---

Training results

Training Loss	Epoch	Step	Validation Loss	F1 Macro	Accuracy
0.7334	1.0	1000	0.2508	0.8941	0.9170
0.2016	2.0	2000	0.1881	0.9096	0.9330
0.1450	3.0	3000	0.1981	0.9119	0.9355
0.1178	4.0	4000	0.2229	0.9158	0.9390
0.0903	5.0	5000	0.2469	0.9161	0.9385
0.0808	6.0	6000	0.2489	0.9100	0.9355

The best model checkpoint was selected based on macro F1 score.

---

Framework versions

• Transformers: 4.57.1
• PyTorch: 2.8.0+cu126
• Datasets: 4.4.1
• Tokenizers: 0.22.1

---

Source code

Training and evaluation code is available on GitHub:
https://github.com/Abdelrahmanemam01/Sentiment-Analysis