RecGPT-V3, a new recommender system from Taobao, addresses challenges in large language model-based recommendations by introducing stateful behavior modeling, a hybrid-modal approach, and efficient reasoning. The system utilizes a Memory Hub to condense user history, reducing computation by over 55%. It also employs a Hybrid-modal Foundation Model that jointly reasons over text tags and Semantic IDs for better item grounding. This new iteration has been deployed in Taobao's "Guess What You Like" feed, showing significant improvements in user engagement and commercial metrics while reducing serving resource consumption by over 50%. AI
IMPACT This advancement in recommender systems could lead to more personalized and efficient user experiences across e-commerce platforms.
RANK_REASON The cluster describes a technical report detailing a new version of a recommender system with performance improvements and new features.
Read on Hugging Face Daily Papers →
- Guess What You Like
- Hybrid-modal Foundation Model
- Memory Hub
- RecGPT-V1
- RecGPT-V2
- RecGPT-V3
- Semantic IDs
- Taobao
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →
