Khyeongjun/bert-based-uncased-agnews4-v01

Model Card: bert-base-uncased-agnews-classifier

Model Details

• Model Name: bert-base-uncased-agnews-classifier
• Base Model: bert-base-uncased
• Model Type: Transformer-based text classification model
• Language: English
• Task: News topic classification (4-way classification)
• Fine-tuned Dataset: AG News
• Developed by: [Your Name or Organization]
• License: Apache 2.0
• Release Date: October 2025

---

Model Description

This model fine-tunes the bert-base-uncased model on the AG News dataset for news topic classification. It classifies English news headlines or short articles into one of four categories: World, Sports, Business, and Science/Technology.

---

Intended Uses and Limitations

Intended Uses

• Automatic categorization of English news articles.
• Topic-based organization of media content.
• Text classification tasks for educational or research purposes.

Limitations

• Designed for English text only.
• May not generalize to domains outside of formal news (e.g., social media posts).
• Does not perform fact verification or sentiment analysis.

---

Training Details

Training Procedure

• Framework: PyTorch
• Library: Hugging Face Transformers
• Base Model: bert-base-uncased
• Optimizer: AdamW
• Learning Rate: 2e-5
• Batch Size: 16
• Epochs: 3
• Max Sequence Length: 128

Hardware

• Fine-tuning performed on a single NVIDIA GPU.

---

Evaluation

Metric	Score
Accuracy	0.86
F1 Score (macro)	0.86

Replace the above with your actual evaluation metrics if available.

---

Dataset Details

• Dataset: AG News (Version: latest)
• Training samples: 120,000
• Test samples: 7,600
• Classes: 4
• License: CC0 1.0 Universal

Label Mapping

Label ID	Category
0	World
1	Sports
2	Business
3	Science/Technology

---

How to Use

from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch

model_name = "your-username/bert-base-uncased-agnews-classifier"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForSequenceClassification.from_pretrained(model_name)

text = "NASA launches new satellite to study climate change."
inputs = tokenizer(text, return_tensors="pt")

with torch.no_grad():
    outputs = model(**inputs)
    probabilities = torch.nn.functional.softmax(outputs.logits, dim=-1)
    predicted_label = torch.argmax(probabilities).item()

labels = ["World", "Sports", "Business", "Science/Technology"]
print(f"Predicted label: {labels[predicted_label]}")

---

Bias, Risks, and Ethical Considerations

• The AG News dataset primarily contains articles from English-language Western media sources, which may introduce geographical and cultural bias.
• The model’s predictions reflect the distribution of topics and wording found in that dataset.
• It should not be used to infer real-world importance, bias, or sentiment toward people or groups.
• Always validate model outputs before deploying in production or decision-making systems.

---

Environmental Impact

Training used a single GPU for a few epochs; estimated carbon footprint is minimal. Fine-tuning on small datasets like AG News is relatively lightweight compared to large-scale pretraining.

---

Citation

If you use this model, please cite:

@misc{yourname2025bertagnews,
  title = {BERT-based AG News Classifier},
  author = {Your Name},
  year = {2025},
  howpublished = {\url{https://huggingface.co/your-username/bert-base-uncased-agnews-classifier}},
}

---

Acknowledgements

• Base model: BERT (Devlin et al., 2018)
• Dataset: AG News Corpus (Zhang et al., 2015)
• Implementation: Hugging Face Transformers