🤖 Keras-TensorFlow Product Classifier

📖 Project Description

Keras-TensorFlow Product Classifier is an automatic product classification system based on product names using modern machine learning methods.

🎯 What the system can do:

• Automatically determines product category from name
• Works with different languages
• Uses modern transformers for understanding meaning
• Achieves high accuracy on properly formatted data
• Processes multiple product categories automatically

🏗️ Technical Architecture:

• Sentence Transformers - for creating name embeddings
• Neural Network - for category classification
• TensorFlow/Keras - modern ML framework
• Scikit-learn - for metrics and data preprocessing

🚀 Quick Start

📋 Requirements

• Python 3.8+
• 4GB+ RAM
• TensorFlow 2.20.0+

⚡ Installation

# Clone repository
git clone <repository-url>
cd ai-product-classifier

# Create virtual environment
python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

# Install dependencies
pip install -r requirements.txt

📊 Data Requirements

🔒 Important Note

This project does not include training data due to commercial confidentiality. Users must provide their own product data.

📋 Required Data Format

CSV File Structure

Your data must be in CSV format with the following columns:

name,fullGroupName
"iPhone 15 Pro Max","Accessories/Phone Cases/Apple iPhone 15 Pro Max"
"Samsung Galaxy S24","Accessories/Phone Cases/Samsung Galaxy S24"
"Apple AirPods Pro","Headphones/Apple/AirPods Pro"

Column Descriptions:

• name (required): Product name/title

  • Can be in different languiges
  • Should be descriptive and clear
  • Example: "iPhone 15 Pro Max silicone case"

• fullGroupName (required): Full category path

  • Hierarchical structure with "/" separators
  • Should be consistent across similar products
  • Example: "Accessories/Phone Cases/Apple iPhone 15 Pro Max"

Data Quality Requirements:

• Minimum 3 products per category (system requirement)
• Consistent naming for similar categories
• No empty values in required columns
• UTF-8 encoding for text support

Recommended Data Size:

• Training: 10,000+ products for good results
• Categories: 100+ different categories
• Balance: At least 5-10 products per category

📁 File Placement

Place your CSV file in the project root directory:

ai-product-classifier/
├── your_products.csv    ← Place your data here
├── main.py
├── testing_model.py
└── requirements.txt

🎯 Usage Examples

Training the model with your data:

# Ensure your CSV file is in the project directory
python main.py

Testing predictions:

python testing_model.py

📊 Expected Results

🎯 Model Performance

With properly formatted data, the system typically achieves:

• Accuracy: 85-95% (depends on data quality)
• Training Time: 15-30 seconds (depends on data size)
• Categories: Automatically detects from your data
• Test Samples: 20% of your data (automatically split)

📈 Training Process

The model automatically:

1. Loads and validates your product data
2. Creates text embeddings using multilingual transformers
3. Trains neural network with early stopping
4. Evaluates performance and saves results

🏗️ Project Structure

keras-tensorflow-product-classifier/
├── main.py                 # Main training script
├── testing_model.py        # Model testing script
├── requirements.txt        # Dependencies
├── plots/                  # Visualization graphs
├── .gitignore             # Git ignore rules
└── README.md              # This file

⚙️ Configuration

Model Parameters

• Embedding Model: paraphrase-multilingual-MiniLM-L12-v2
• Neural Network: 384 → 192 → output neurons
• Regularization: L2 + Dropout
• Optimizer: AdamW with learning rate scheduling

Data Processing

• Train/Test Split: 80%/20%
• Stratification: Preserves category proportions
• Minimum Category Size: 3+ products per category

📈 Performance Analysis

The system provides detailed classification reports including:

• Per-category precision, recall, and F1-score
• Confusion matrix analysis
• Training history visualization
• Error analysis for difficult categories

📊 Visualization Results

Training Progress

Model accuracy and loss during training epochs

Training Time Breakdown

Time distribution across training stages

Top Categories

Best performing product categories

Confusion Matrix

Classification performance heatmap

🚀 Future Improvements

• Web interface for easy testing
• Real-time classification API
• Support for new product categories
• Performance optimization
• Multi-language support expansion

📝 License

License Type: Open Source / Free Software

This project is created for educational and demonstration purposes and is available under open source terms.

---

Note: This system demonstrates modern ML techniques for product classification. Users must provide their own training data in the specified format.