metadata

title: News Classification
emoji: 📰
colorFrom: green
colorTo: indigo
sdk: docker
sdk_version: 0.95.2
app_file: app.py
pinned: false
language: en
license: mit
tags:
  - text-classification
  - news-classification
  - LSTM
  - tensorflow
pipeline_tag: text-classification

News Source Classifier

This model classifies news headlines as either Fox News or NBC News using a deep learning LSTM (Long Short-Term Memory) neural network architecture.

Model Description

Architecture

Input Layer: Embedding layer (vocab_size=74,934, embedding_dim=128)
LSTM Layer 1: 128 units with return sequences
Dropout Layer 1: For regularization
LSTM Layer 2: 64 units
Dropout Layer 2: For regularization
Output Layer: Dense layer with 2 units (binary classification)

Technical Details

Total Parameters: 9,772,676 (37.28 MB)
Training Parameters: 9,772,674 (37.28 MB)
Input Shape: (41, ) - sequences of length 41
Performance: Achieves binary classification of news sources

Usage

You can use this model through our REST API:

import requests

def predict_news_source(text):
    response = requests.post(
        "https://jiahuita-news-classification.hf.space/predict",
        json={"text": text},
        headers={"Content-Type": "application/json"}
    )
    return response.json()

# Example usage
headline = "Scientists discover breakthrough in renewable energy research"
result = predict_news_source(headline)
print(result)

Example response:

{
    "label": "nbc",
    "score": 0.789
}

Limitations and Bias

This model has been trained on news headlines from specific sources and time periods, which may introduce certain biases:

Training data is limited to two news sources
Headlines represent a specific time period
Model may be sensitive to writing style rather than just content

Training Details

The model was trained using:

TensorFlow 2.10.0
Binary cross-entropy loss
Embedding layer for text representation
Dual LSTM layers with dropout for robust feature extraction
Dense layer with softmax activation for final classification

License

This project is licensed under the MIT License.