README.md

metadata

title: ACE Classifier Doc2Vec
emoji: 🤖
colorFrom: blue
colorTo: green
sdk: custom
app_port: 8080

ACE Content Attribution Classifier (Doc2Vec)

This model classifies content as either "attributed" or "unattributed" using Doc2Vec embeddings and machine learning classifiers.

Model Details

• Training Date: 2025_09_17
• Architecture: Doc2Vec + Machine Learning Classifier
• Task: Binary text classification
• Classes: attributed, unattributed

Usage

API Format

Send POST requests to the inference endpoint:

{
  "inputs": {
    "content": "Your content text here",
    "meta_description": "Optional meta description"
  }
}

Response Format

[
  {
    "label": "attributed",
    "score": 0.75
  },
  {
    "label": "unattributed",
    "score": 0.25
  }
]

Python Example

import requests

api_url = "https://api-inference.huggingface.co/models/athenahq/ACE-classifier-doc2vec"
headers = {"Authorization": "Bearer YOUR_HF_TOKEN"}

data = {
    "inputs": {
        "content": "Machine learning models for content attribution analysis",
        "meta_description": "A comprehensive guide to ML-based content classification"
    }
}

response = requests.post(api_url, headers=headers, json=data)
result = response.json()
print(result)

cURL Example

curl -X POST \
  https://api-inference.huggingface.co/models/athenahq/ACE-classifier-doc2vec \
  -H "Authorization: Bearer YOUR_HF_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "inputs": {
      "content": "Your content text here",
      "meta_description": "Optional meta description"
    }
  }'

Model Performance

The model uses the best-performing combination from extensive hyperparameter tuning across multiple Doc2Vec configurations and classifiers.

Files

• handler.py: Custom inference handler
• model_summary.json: Overview of all trained models
• rank_1_*_classifier.pkl: Best performing classifier
• rank_1_*_doc2vec.model: Best performing Doc2Vec model
• rank_1_*_metadata.json: Model metadata and configuration

Technical Details

• Doc2Vec: Uses both PV-DM and PV-DBOW algorithms
• Preprocessing: Text cleaning, tokenization, and filtering
• Classifiers: Random Forest, SVM, Logistic Regression, Neural Networks
• Evaluation: Comprehensive accuracy and confidence analysis