README.md

---
license: mit
library_name: transformers
pipeline_tag: text-classification
tags:
- transformers
- pytorch
- text-classification
- news-classification
- ag-news
- nlp
---

# News Classification using RoBERTa

## Model Description

This model is a fine-tuned RoBERTa model for multiclass news classification. It predicts the category of a news headline or article among four classes:

* World
* Sports
* Business
* Sci/Tech

The model was fine-tuned on the AG News dataset using Hugging Face Transformers and PyTorch.

### Model Details

* **Developed by:** MhoOmm
* **Model Type:** RoBERTa for Sequence Classification
* **Language:** English
* **License:** MIT
* **Finetuned From:** roberta-base

## Labels

| Label ID | Category |
| -------- | -------- |
| 0        | World    |
| 1        | Sports   |
| 2        | Business |
| 3        | Sci/Tech |

## Intended Use

### Direct Use

This model can be used for:

* News article categorization
* Content organization systems
* News recommendation pipelines
* NLP learning projects
* Text classification demonstrations

### Out-of-Scope Use

This model is not intended for:

* Medical diagnosis
* Financial advice
* Legal decision making
* Safety-critical systems
* Classification of non-English news

## Usage

```python
from transformers import AutoTokenizer
from transformers import AutoModelForSequenceClassification
import torch

model_name = "MhoOmm/news-classifier"

tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForSequenceClassification.from_pretrained(model_name)

text = "Apple launches a new AI-powered processor"

inputs = tokenizer(
    text,
    return_tensors="pt",
    truncation=True
)

with torch.no_grad():
    outputs = model(**inputs)

prediction = outputs.logits.argmax(-1).item()

print(model.config.id2label[prediction])
```

## Training Data

The model was fine-tuned on the AG News dataset.

### Dataset Statistics

* Training Samples: 120,000
* Test Samples: 7,600
* Classes: 4

### Categories

* World
* Sports
* Business
* Sci/Tech

## Training Procedure

### Preprocessing

* RoBERTa tokenizer
* Maximum sequence length: 128
* Truncation enabled
* Dynamic padding

### Hyperparameters

* Learning Rate: 2e-5
* Batch Size: 16
* Epochs: 3
* Optimizer: AdamW
* Framework: PyTorch
* Trainer: Hugging Face Trainer

## Evaluation

### Metrics

The model was evaluated using classification accuracy.

### Example Predictions

| Input                                         | Prediction |
| --------------------------------------------- | ---------- |
| India defeats Australia in the cricket final  | Sports     |
| Stock market reaches a new all-time high      | Business   |
| Apple unveils a new AI-powered device         | Sci/Tech   |
| Global leaders meet to discuss climate policy | World      |

### Results

* Accuracy: **94.76%**

## Limitations

* Trained only on English-language news articles.
* May perform poorly on informal text such as tweets and social media posts.
* Restricted to the four AG News categories.
* Performance may decrease on domains significantly different from the training data.

## Architecture

* Base Model: RoBERTa Base
* Hidden Size: 768
* Transformer Layers: 12
* Attention Heads: 12
* Classification Head: Linear Layer (768 → 4)

## Software

* PyTorch
* Transformers
* Hugging Face Hub
* Gradio

## Author

Created by **MhoOmm** as an end-to-end NLP project demonstrating:

* Transformer fine-tuning
* Text classification
* Model deployment with Hugging Face
* Interactive inference using Gradio

## Citation

If you use this model, please cite the original RoBERTa paper:

**RoBERTa: A Robustly Optimized BERT Pretraining Approach**

Yinhan Liu, Myle Ott, Naman Goyal, et al.

https://arxiv.org/abs/1907.11692