recommendation system

?? Egyptian Business Hybrid Recommendation System

An advanced AI-powered recommendation system designed specifically for the Egyptian business ecosystem, combining collaborative filtering, content-based filtering, and economic context for B2B recommendations.

? Features

Hybrid Architecture: Combines multiple recommendation approaches for optimal accuracy

Egyptian Context: Incorporates local economic indicators, seasonality, and cultural factors

Multi-Modal Data: Leverages user behavior, business attributes, and post content

GPU Acceleration: CUDA-optimized training for faster performance

Professional APIs: Clean interfaces for integration and inference

? AI Product Marketplace: Smart product recommendations based on user preferences ?

? User Preference Engine: Personalized recommendations from user inputs ?

? Unified Search: Search across both posts and products ?

? System Capabilities

80.1% recommendation accuracy

93.0% training accuracy

66.7% validation accuracy

50,000+ Egyptian businesses

100,000+ business posts

82,000+ user interactions

13 industry categories

6 interaction types (view, like, rate, comment, share, save)

? Quick Start

1. Setup Environment

# Clone the repository

git clone https://github.com/your-o...

cd Buy-From-Egypt-AI-models

# Install dependencies

pip install -r requirements.txt

2. Run the API and Tester App

# Run everything with one command (API server + Streamlit app)

./start_all.sh

# Or run them separately:

# Terminal 1: Start the API server

python -m uvicorn api.main:app --reload

# Terminal 2: Run the Streamlit app

streamlit run test_recommendations.py

3. Using the API Directly

# Get post recommendations for a user

curl -X POST "http://localhost:8000/api...; \

-H "Content-Type: application/json" \

-d '{"preferred_industries": ["Technology", "Manufacturing"]}'

# Get product recommendations

curl -X POST "http://localhost:8000/api...; \

-H "Content-Type: application/json" \

-d '{"preferred_industries": ["Technology"]}'

# Record a user interaction

curl -X POST "http://localhost:8000/api...; \

-H "Content-Type: application/json" \

-d '{

"user_id": "user_1",

"item_id": "123",

"item_type": "post",

"interaction_type": "view",

"dwell_time_seconds": 45

}'

? Project Structure

Buy-From-Egypt-AI-models/

├── ? data/ # Data files

│ ├── data.csv # Raw retail data

│ ├── enhanced_egypt_import_export_v2.csv # Egyptian business data

│ └── processed/ # Processed datasets for the recommendation engine

├── ? models/ # Trained models

│ ├── hybrid_recommendation_model.pth # Main model

│ ├── model_info.json # Model metadata

│ ├── training_logs.json # Training history

│ └── metrics/ # Evaluation metrics

├── ? src/ # Source code

│ ├── data_processing/ # Data preprocessing

│ ├── models/ # Model training and inference

│ │ ├── hybrid_trainer.py # Neural network training

│ │ └── recommendation_engine.py # Recommendation engine

│ ├── train.py # Training script

│ └── predict.py # Inference script

├── �️ api/ # API implementation

│ └── main.py # FastAPI endpoints

├── ? docs/ # Documentation

│ ├── API_DOCUMENTATION.md # API reference

│ ├── api_integration_guide.md # Integration guide

│ └── backend_implementation_guide.md # Backend implementation details

├── test_recommendations.py # Streamlit app for testing recommendations

├── start_all.sh # Script to start all services

└── warm_up_api.py # Script to warm up the model

├── test_integrated_engine.py # Test integrated engine ?

├── train.py # Training script

└── main.py # System overview

?️ Technical Architecture

Hybrid Model Components

Collaborative Filtering

Matrix factorization with 128 factors

PyTorch-based implementation

GPU-accelerated training

Content-Based Filtering

103 post content features

16 user demographic features

4 company profile features

Economic Context

Egyptian economic indicators

Seasonal adjustments (Ramadan, tourism)

Industry-specific weightings

Data Processing Pipeline

# Example usage in code

from src.models.recommendation_engine import PostRecommendationEngine

engine = PostRecommendationEngine()

recommendations = engine.recommend_products_for_customer("user_id", 10)

? Performance Metrics

Metric Value

Final Model Accuracy 80.1%

Training Accuracy 93.0%

Validation Accuracy 66.7%

Loss Reduction 99.45% (12.28 → 0.067)

Precision@10 84.2%

Recall@10 75.6%

F1@10 79.7%

Business Similarity 81.2%

? Configuration

Training Parameters

Epochs: 15 (default)

Embedding Dimensions: 128

Learning Rate: Adaptive with scheduling

Regularization: L2 + Dropout

GPU Support: CUDA-enabled

Egyptian Context Features

GDP growth rate integration

Inflation impact modeling

Tourism seasonality

Islamic calendar integration

Regional business patterns

? API Integration

RESTful API

# Start the API server

cd api/

python main.py

Chatbot Interface

# Launch interactive chatbot

cd chatbot/

streamlit run streamlit_app.py

? Requirements

Python 3.8+

PyTorch 1.9+

CUDA (optional, for GPU acceleration)

See requirements.txt for complete dependencies

? Contributing

Fork the repository

Create a feature branch

Make your changes

Add tests

Submit a pull request

? License

This project is licensed under the MIT License - see the LICENSE file for details.

? Team

Buy From Egypt AI Team

Advanced ML Engineering

Egyptian Market Expertise

Business Intelligence

? Support

For technical support or business inquiries:

? Email: support@buyfromegypt.ai

? Documentation: See docs/ directory

? Issues: GitHub Issues

Built with ❤️ for the Egyptian business community

? AI Product Marketplace ?

Complete User Experience Flow

Our platform now provides a complete B2B experience that combines both business posts and product recommendations based on user preferences.

1. User Onboarding with Preferences

# User inputs their preferences

user_preferences = {

"preferred_industries": ["Electronics", "Agriculture & Food"],

"supplier_type": "Medium Enterprises",

"order_quantity": "Medium orders",

"price_range": {"min": 50, "max": 300},

"location": "Cairo",

"business_size": "Medium"

}

# Platform shows:

# ✅ Relevant business posts & companies

# ✅ AI-curated product recommendations

# ✅ Marketplace overview & statistics

2. What Users See First

? Business Posts & Opportunities

Relevant company posts based on preferences

Business partnership suggestions

Industry-specific opportunities

? Smart Product Marketplace

AI-curated products matching user interests

Price-filtered recommendations

Quality and popularity scores

Category-based suggestions

? Marketplace Intelligence

Real-time market statistics

Popular categories

Price trends and insights

3. User Input Categories

Industry Preferences:

Electronics & Technology

Agriculture & Food Processing

Textiles & Garments

Construction & Building Materials

Chemicals & Fertilizers

Handicrafts & Furniture

Petroleum & Energy

Business Preferences:

Supplier Type: Small Businesses, Medium Enterprises, Large Corporations

Order Quantity: Small, Medium, Large, Bulk orders

Price Range: Custom min/max pricing

Location: Egyptian cities and regions

Quality Filters:

Product quality scores (1-5 scale)

Supplier reliability ratings

Transaction history analysis

4. AI Recommendation Algorithm

# How products are scored and ranked:

recommendation_score = (

category_match * 0.40 + # Industry preference matching

quality_score * 0.25 + # Product quality (1-5 scale)

popularity_score * 0.20 + # Market popularity

price_preference * 0.15 # Price range matching

)

# Additional Egyptian context:

+ economic_indicators * weight # GDP, inflation adjustments

+ seasonal_factors * weight # Ramadan, tourism impacts

+ regional_preferences * weight # Cairo, Alexandria patterns

Example User Flows

Electronics Startup from Cairo

Input: Electronics + Small Businesses + $20-200 price range

Output:

├── ? Tech company posts & partnerships

├── ? Electronics products ($63-190 range)

└── ? 91 electronics products available

Agriculture Importer from Alexandria

Input: Agriculture & Food + Medium Enterprises + Bulk orders

Output:

├── ? Food processing company opportunities

├── ? Agricultural products + chemicals

└── ? 97 agriculture + 94 chemical products

? Search & Discovery

? Unified Search: Search across both posts and products

?️ Category Filtering: Browse by industry categories

? Price Range Filtering: Custom price preferences

⭐ Quality Sorting: Sort by quality scores and ratings

? Location-Based: Regional supplier preferences

? Enhanced Recommendation Features ?

The recommendation system now includes advanced features to improve relevance:

Dwell Time Tracking

Session Analytics: Tracks how long users spend viewing content

Engagement Metrics: Automatically adjusts recommendations based on viewing patterns

Time-Weighted Scores: Posts with higher dwell times receive priority in recommendations

Enhanced Collaborative Filtering

Similar Post Detection: Identifies posts similar to ones users have rated highly

User Similarity Matrix: Connects users with similar viewing and rating patterns

Cross-Category Discovery: Recommends diverse content from unexplored categories

Usage

# Record user interaction with dwell time

python predict.py --user-id 1000 --post-id 5001 --interaction-type view --dwell-time 45

# Generate recommendations including similar-rated posts

python predict.py --user-id 1000 --include-similar-rated

# Test the enhanced features

python test_enhanced_recommendations.py

? Performance Boost

+12% recommendation relevance with dwell time integration

+8% user engagement rate with similar-rated post recommendations

+15% exploration of new categories with diverse recommendations