Scalable Online Deep Learning for Stock Recommendations

A recent research article introduces a scalable online deep learning framework designed for stock recommendations, tackling issues related to concept drift and the need for low-latency predictions. This system is built on a distributed microservices architecture utilizing Kubernetes, Docker, and RabbitMQ. It features a hybrid leader-follower approach, where a primary model is continuously trained on real-time financial data (EPS, MACD, price) sourced from the Alpha Vantage API, while parallel replica models provide recommendations. Additionally, a multilayer perceptron, developed with TensorFlow Recommenders, creates content-based suggestions based on explicit user ratings (1-5) and employs transfer learning. The architecture guarantees high availability and resilience against faults.