| # 🧠 BERT-based 4-Class News Classifier |
|
|
| A fine-tuned [BERT-base-uncased](https://huggingface.co/bert-base-uncased) model for classifying news articles into four categories: **World**, **Sport**, **Business**, and **Science & Technology**. |
|
|
| ## 📖 Model Overview |
|
|
| This model adapts BERT for multi-class sequence classification. |
| It was trained on a labeled news dataset with four categories and achieves strong performance on short text snippets such as news headlines or article summaries. |
|
|
| | Label ID | Category | |
| |-----------|-----------| |
| | 0 | World | |
| | 1 | Sport | |
| | 2 | Business | |
| | 3 | Science & Technology | |
|
|
| ## ⚙️ Training Details |
|
|
| - **Base model:** `bert-base-uncased` |
| - **Task:** Sequence classification (4 classes) |
| - **Training environment:** Google Colab (T4 GPU) |
| - **Max sequence length:** 200 |
| - **Batch size:** 8 |
| - **Epochs:** 3 |
| - **Optimizer:** AdamW (lr=2e-5) |
| - **Framework:** PyTorch + 🤗 Transformers Trainer API |
|
|
| The dataset was tokenized using `AutoTokenizer` with truncation and dynamic padding enabled. |
| Model and tokenizer were saved using `save_pretrained()` and uploaded to the Hugging Face Hub. |
|
|
| ## 🧩 Usage |
|
|
| You can load and use the model directly with the 🤗 Transformers library: |
|
|
| ```python |
| from transformers import AutoTokenizer, AutoModelForSequenceClassification |
| import torch |
| |
| # Load model from the Hub |
| model_name = "zacanadir/wsbt-sequence-classifier" |
| tokenizer = AutoTokenizer.from_pretrained(model_name) |
| model = AutoModelForSequenceClassification.from_pretrained(model_name) |
| |
| MAX_LENGTH = 200 |
| |
| def predict_text(text_list): |
| encodings = tokenizer( |
| text_list, |
| padding=True, |
| truncation=True, |
| max_length=MAX_LENGTH, |
| return_tensors="pt" |
| ).to(model.device) |
| |
| model.eval() |
| with torch.no_grad(): |
| outputs = model(**encodings) |
| preds = torch.argmax(outputs.logits, dim=-1) |
| return preds.tolist() |
| |
| # Example |
| sample_texts = [ |
| "Nvidia becomes the first chipmaker to top $5 trillion in market value.", |
| "The Raja of Casablanca advances to the semi final of the African Champions' League." |
| ] |
| |
| label_to_category = { |
| 0: "World", |
| 1: "Sport", |
| 2: "Business", |
| 3: "Science & Technology" |
| } |
| |
| preds = predict_text(sample_texts) |
| predicted_categories = [label_to_category[int(i)] for i in preds] |
| print(predicted_categories) |
| |
| 🧾 Output Example |
| ['Business', 'Sport'] |
| |
| 🚀 How to Reuse / Fine-tune Further |
| |
| If you’d like to fine-tune this model on your own dataset: |
| |
| from transformers import Trainer, TrainingArguments, AutoTokenizer, AutoModelForSequenceClassification |
| from datasets import load_dataset |
| |
| tokenizer = AutoTokenizer.from_pretrained("zacanadir/wsbt-sequence-classifier") |
| model = AutoModelForSequenceClassification.from_pretrained("zacanadir/wsbt-sequence-classifier", num_labels=4) |
| |
| # Load your dataset and fine-tune |
| |
| 📦 Files Included |
| File Description |
| config.json Model architecture & label mapping |
| pytorch_model.bin Trained weights |
| tokenizer.json, vocab.txt, special_tokens_map.json Tokenizer files |
| training_args.bin Trainer configuration (epochs, LR, etc.) |
| |
| ✨ Acknowledgements |
| Hugging Face Transformers |
| Google Colab for free GPU training |
| PyTorch |
| |
| Trained and deployed with ❤️ using Hugging Face and Google Colab. |
