File size: 2,987 Bytes
e4144f4 c6715fc e4144f4 c6715fc e4144f4 eebd529 e4144f4 764e674 e4144f4 c6715fc e4144f4 764e674 e4144f4 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 |
---
license: mit
language:
- en
metrics:
- accuracy
- bertscore
- f1
base_model:
- google-bert/bert-base-uncased
pipeline_tag: text-classification
---
# Newswire Classifier (AP, UPI, NEA) - BERT Transformers
## 📘 Overview
This repository contains three separately trained BERT models for identifying whether a newspaper article was produced by one of three major newswire services:
- **AP (Associated Press)**
- **UPI (United Press International)**
- **NEA (Newspaper Enterprise Association)**
The models are designed for historical news classification from public-domain newswire articles (1960–1975).
## 🧠 Model Architecture
- **Base Model:** `bert-base-uncased`
- **Task:** Binary classification (`1` if from the specific newswire, `0` otherwise)
- **Optimizer:** AdamW
- **Loss Function:** Binary Cross-Entropy with Logits
- **Batch Size:** 16
- **Epochs:** 4
- **Learning Rate:** 2e-5
- **Device:** TPU (v2-8) in Google Colab
## 📊 Training Data
- **Source:** Historical newspapers (1960–1975, public domain)
- **Articles:** 4000 per training round (1000 from target newswire, 3000 from other sources)
- **Features Used:** Headline, author, and first 100 characters of the article.
- **Labeling:** `1` for articles from the target newswire, `0` for all others.
## 🚀 Model Performance
| Model | Accuracy | Precision | Recall | F1 Score |
|-------|----------|----------|-------|----------|
| **AP** | 0.9925 | 0.9926 | 0.9925 | 0.9925 |
| **UPI** | 0.9999 | 0.9999 | 0.9999 | 0.9999 |
| **NEA** | 0.9875 | 0.9880 | 0.9875 | 0.9876 |
## 🛠️ Usage
### Installation
```bash
pip install transformers torch
```
### Example Inference (AP Classifier)
```python
from transformers import AutoModelForSequenceClassification, AutoTokenizer
model = AutoModelForSequenceClassification.from_pretrained("mike-mcrae/newswire_classifier/AP")
tokenizer = AutoTokenizer.from_pretrained("mike-mcrae/newswire_classifier/AP")
text = "(AP) President speaks at conference..."
inputs = tokenizer(text, return_tensors="pt", truncation=True, max_length=128)
outputs = model(**inputs)
prediction = outputs.logits.argmax().item()
print("AP Article" if prediction == 1 else "Not AP Article")
```
## ⚙️ Recommended Usage Notes
- The models were trained on a combination of the first 100 characters of headline + author + the first 100 characters of articles, as the mention of the newswire often appears in these sections. Using the same format for inference may improve accuracy.
## 📜 Licensing & Data Source
- **Training Data:** Historical newspaper articles (1960–1975) from public-domain sources.
- **License:** Public domain (for data) and MIT License (for model and code).
## 💬 Citation
If you use these models, please cite:
```
@misc{newswire_classifier,
author = {McRae, Michael},
title = {Newswire Classifier (AP, UPI, NEA) - BERT Transformers},
year = {2025},
publisher = {Hugging Face},
url = {https://huggingface.co/username/newswire_classifier}
}
``` |