Researchers have developed CAP-Mix, a novel generative retrieval method for recommender systems that represents items as a soft mixture of shared prototypes and item-specific vectors. This approach aims to retain more information than discrete Semantic IDs while introducing structure absent in flat continuous vector tables. Experiments show CAP-Mix outperforms existing methods on sparse Amazon datasets and offers significant gains in HR@5 and NDCG@5 across multiple benchmarks, while also providing much higher inference throughput compared to discrete baselines. AI
IMPACT This research could lead to more efficient and accurate recommender systems by improving item representation techniques.
RANK_REASON The cluster contains an academic paper detailing a new method for recommender systems.
- alphaXiv
- arXiv
- CAP-Mix
- CatalyzeX
- CORE Recommender
- DagsHub
- Gotit.pub
- Hugging Face
- Influence Flower
- ScienceCast
- Semantic IDs
- Steam
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →