Generating Long Semantic IDs in Parallel for Recommendation

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Hou, Yupeng, Li, Jiacheng, Shin, Ashley, Jeon, Jinsung, Santhanam, Abhishek, Shao, Wei, Hassani, Kaveh, Yao, Ning, McAuley, Julian
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866916781958889472
author Hou, Yupeng
Li, Jiacheng
Shin, Ashley
Jeon, Jinsung
Santhanam, Abhishek
Shao, Wei
Hassani, Kaveh
Yao, Ning
McAuley, Julian
author_facet Hou, Yupeng
Li, Jiacheng
Shin, Ashley
Jeon, Jinsung
Santhanam, Abhishek
Shao, Wei
Hassani, Kaveh
Yao, Ning
McAuley, Julian
contents Semantic ID-based recommendation models tokenize each item into a small number of discrete tokens that preserve specific semantics, leading to better performance, scalability, and memory efficiency. While recent models adopt a generative approach, they often suffer from inefficient inference due to the reliance on resource-intensive beam search and multiple forward passes through the neural sequence model. As a result, the length of semantic IDs is typically restricted (e.g. to just 4 tokens), limiting their expressiveness. To address these challenges, we propose RPG, a lightweight framework for semantic ID-based recommendation. The key idea is to produce unordered, long semantic IDs, allowing the model to predict all tokens in parallel. We train the model to predict each token independently using a multi-token prediction loss, directly integrating semantics into the learning objective. During inference, we construct a graph connecting similar semantic IDs and guide decoding to avoid generating invalid IDs. Experiments show that scaling up semantic ID length to 64 enables RPG to outperform generative baselines by an average of 12.6% on the NDCG@10, while also improving inference efficiency. Code is available at: https://github.com/facebookresearch/RPG_KDD2025.
format Preprint
id arxiv_https___arxiv_org_abs_2506_05781
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Generating Long Semantic IDs in Parallel for Recommendation
Hou, Yupeng
Li, Jiacheng
Shin, Ashley
Jeon, Jinsung
Santhanam, Abhishek
Shao, Wei
Hassani, Kaveh
Yao, Ning
McAuley, Julian
Information Retrieval
Semantic ID-based recommendation models tokenize each item into a small number of discrete tokens that preserve specific semantics, leading to better performance, scalability, and memory efficiency. While recent models adopt a generative approach, they often suffer from inefficient inference due to the reliance on resource-intensive beam search and multiple forward passes through the neural sequence model. As a result, the length of semantic IDs is typically restricted (e.g. to just 4 tokens), limiting their expressiveness. To address these challenges, we propose RPG, a lightweight framework for semantic ID-based recommendation. The key idea is to produce unordered, long semantic IDs, allowing the model to predict all tokens in parallel. We train the model to predict each token independently using a multi-token prediction loss, directly integrating semantics into the learning objective. During inference, we construct a graph connecting similar semantic IDs and guide decoding to avoid generating invalid IDs. Experiments show that scaling up semantic ID length to 64 enables RPG to outperform generative baselines by an average of 12.6% on the NDCG@10, while also improving inference efficiency. Code is available at: https://github.com/facebookresearch/RPG_KDD2025.
title Generating Long Semantic IDs in Parallel for Recommendation
topic Information Retrieval
url https://arxiv.org/abs/2506.05781