Continuous soft item representations improve recommendation accuracy and speed
Item Representations Matter: Bridging Hard Semantic IDs and Soft Prototype Mixtures for Generative Retrieval
Information RetrievalMachine Learning
Summary
Recommender systems suggest items like products or games that you might like next. Current methods either use strict categories that lose some item details or treat each item separately without sharing information, which isn’t always efficient. The authors propose a new method called CAP-Mix that represents each item as a mix of shared prototypes plus a unique part, keeping more information. This approach improves recommendation accuracy and speeds up the process compared to existing methods, especially on sparse datasets like Amazon product data.
What this means in practice
- •For ecommerce platform developers: Improve product recommendation accuracy and speed by using shared prototype mixtures instead of independent item lookups.
- •For digital content curators: Enhance content suggestion systems by representing items as soft combinations of prototypes for better relevance in sparse catalogs.
- •For music streaming service engineers: Design music recommendation features that gain better hit rates and faster results by using continuous soft item embeddings.$Commercial implications: Enables faster and more accurate music recommendations, improving user experience and retention, which can be packaged as a premium feature.
Authors
Vicente Balmaseda, Dongqi Fu, Jiarui Feng, Ren chen, Hanqing Zeng, Tianxin Wei, Yinglong Xia, Tianbao Yang, Hong Li
Abstract
Modern recommender systems rely on generative retrieval techniques to predict the next item either as discrete codes or as a continuous vector. On the one hand, Semantic IDs (SIDs) give discrete models shared parameters and pretrained semantic structure, but discretizing items into hard codes discards some information from the original content embeddings. On the other hand, continuous models avoid quantization, yet typically use flat item lookup tables with no explicit parameter tying across items. We ask whether a continuous recommender can introduce shared structure without making it a discrete bottleneck. We propose CAP-Mix, a flow recommender that represents each item as a soft mixture of shared prototypes plus an item-specific vector. The assignments, prototypes, and item vectors are trained jointly, and the soft representation is retained at inference. Pretrained information is optional: SID responsibilities can serve as a prior over assignments, while a full content embedding can initialize the item-specific vector without changing the initial flat table. In matched comparisons on three datasets, CAP-Mix improves over a flat table on the two sparse Amazon datasets and yields a small gain on Steam when trained without pretrained information. Given only an offline SID, it outperforms a hard table built from the same tokenizer by 4.45-14.90% in HR@5. Given the full content embedding, it matches or improves a strong flat table while hard assignments regress on the sparse Amazon datasets. Overall, averaged across six benchmarks, CAP-Mix improves HR@5 and NDCG@5 by 6.9% and 13.6% relative to the best evaluated baseline for each dataset and metric, while delivering 1.4-161x the inference throughput of the discrete baselines in our experiments.