A Comprehensive Architecture for Federated Learning-Based Smart Advertising.

Area of Science:

• Computer Science
• Artificial Intelligence
• Data Science

Background:

• Smart advertising demands efficient, secure, and private data handling.
• Federated learning (FL) offers a privacy-preserving approach but faces challenges in data integration and resource optimization.
• Existing architectures struggle with scalability and performance in dynamic user environments.

Purpose of the Study:

• To introduce a novel data architecture for smart advertising.
• To leverage federated learning (FL) for enhanced data privacy, integrity, and efficiency.
• To demonstrate significant reductions in network traffic and CPU usage while maintaining advertising accuracy.

Main Methods:

• Developed a data architecture with semi-random role assignment for model, data, and validator nodes.
• Implemented a selective node engagement strategy for optimized resource utilization.
• Utilized federated learning (FL) as the core methodology for data processing and model training.
• Validated the architecture on the AROUND social network platform through simulations and real-world implementation.

Main Results:

• Achieved over 50% reduction in network traffic and average CPU usage.
• Demonstrated sustained FL model accuracy despite resource optimization.
• Showcased scalability with a 20-fold increase in user count.
• Confirmed no negative impact on smart advertising accuracy, click rates, or user engagement.

Conclusions:

• The proposed data architecture effectively balances performance, security, and privacy in smart advertising.
• Federated learning (FL) integration enables efficient and scalable data processing.
• The architecture offers a significant advancement for digital marketing and FL applications.