This is an extensible recommendation system that has evolved over many years of recommendation algorithm development, becoming a relatively mature technical field. While recommendation systems are widely used in large companies, building a recommendation system still poses significant challenges for small and medium-sized enterprises. The overall complexity of the system and the limited customization offered by recommendation system service providers in the market may not meet the specific functional needs of businesses.
To address this issue, I have distilled a relatively universal, miniaturized, and extensible recommendation system from my years of experience in the recommendation system industry. The system aims to help small and medium-sized enterprises build their own recommendation systems quickly and cost-effectively, meeting their specific business requirements.
Versatility: Suitable for various recommendation scenarios, including product recommendations, content recommendations, and more.
Compact Scale: The system is designed with simplicity and lightness, reducing maintenance and operating costs.
Scalability: Flexible expansion and customization according to business needs to meet the personalized requirements of different enterprises.
Recall: Integrates multiple recall methods, including KV recall, vector recall, DSSM vector recall, and DeepMatch vector recall and inference.
Ranking: Utilizes Alibaba Cloud's EAS inference service for ranking.
Re-ranking Strategy: Abstract re-ranking functionality with features such as diversity, forced insertion, weight adjustment, and supports the scalability of strategies.
Read History Filtering: Integrates read history filtering functionality.
Session Filtering: Incorporates session filtering functionality.
AB Test: Comes with AB test traffic segmentation and allocation functionality modules.
It also includes interfaces and functional components for profile acquisition, feature acquisition, and more.
This project is under development, and your contributions are welcome! Contact by [email protected] if you have any question.