From Visual Search to Semantic Recommendations: Visidea, a Comprehensive Recommender System Platform for E-commerce

This research addresses the challenges of developing a scalable, multi-faceted recommendation system that integrates advanced machine learning techniques with practical industrial applications. Visidea incorporates a range of recommendation strategies, from traditional collaborative filtering to cutting-edge visual and semantic search capabilities. The platform’s modular architecture enables easy integration with existing e-commerce systems through a plugin-based approach. Key contributions include: (1) A novel visual search module utilizing YOLO and open-source CLIP models for object detection and embedding; (2) A semantic search functionality allowing natural language queries; (3) An automatic query generation system to address cold start problems and enhance search autocomplete; (4) A robust cloud infrastructure initially developed on AWS and later migrated to GCP, featuring a Kubernetes cluster for efficient batch processing and scalable deployment. This research demonstrates the successful application of advanced AI techniques in a real-world industrial setting, bridging the gap between academic research and practical implementation in the e-commerce domain, while also highlighting the challenges and benefits of cloud platform migration in a production environment