Semantic ID Recommendation

Follow this topic Ask about this topic

Semantic identifier recommendation studies how to assign items compact discrete codes that capture useful content, behavioral, or collaborative relationships. Systems use methods such as residual vector quantization, graph partitioning, and multimodal encoders to build codes, then predict them for sequential recommendation or generative retrieval. The codes can reduce serving costs and help with cold-start items, but collisions, lost fine-grained item distinctions, and limited reachability during generation remain challenges. Current work explores better code allocation and decoding while balancing semantic organization, collaborative accuracy, and production efficiency.

Latest papers

38 papers7 papers in the last four weeks, up 133% on the four weeks before. 0.0% of all new papers.

Conditional Memory Enhanced Item Representation for Generative Recommendation
May 12, 2026Ziwei Liu, Yejing Wang, Shengyu Zhou +2Semantic ID Recommendation Representation Learning
City University of Hong Kong Hong Kong, China · Independent Researcher Beijing, China · Tsinghua University Beijing, China
UxSID: Semantic-Aware User Interests Modeling for Ultra-Long Sequence
May 9, 2026Hongwei Zhang, Qiqiang Zhong, Junfeng Shu +8Semantic ID Recommendation Sequential Recommendation
Kuaishou Technology, Beijing, China
Intent-Driven Semantic ID Generation for Grounded Conversational News Recommendation
May 8, 2026Hongyang Su, Beibei Kong, Lei Cheng +3Semantic ID Recommendation LLM Grounding
Platform and Content Group, Tencent, Shenzhen, China · Shenzhen Campus of Sun Yat-sen University, Shenzhen, China
Beyond Item IDs: Scaling Short-Form-Video Recommendation via Semantic-Native Long Sequence Modeling
May 4, 2026Ruixiao Sun, Diego Uribe Mora, Zhimeng Jiang +7Semantic ID Recommendation Sequential Recommendation
Google Mountain View, USA
Deep Interest Mining for Intent-Enriched Semantic IDs in Multimodal Generative Recommendation
Mar 3, 2026Yangchen Zeng, Jinze WangRecommender Systems Semantic ID Recommendation
Stop Treating Collisions Equally: Qualification-Aware Semantic ID Learning for Recommendation at Industrial Scale
Feb 28, 2026Zheng Hu, Yuxin Chen, Yongsen Pan +13Semantic ID Recommendation Multimodal Recommendation
University of Electronic Science and Technology of China, Chengdu, China · Kuaishou Technology, Beijing, China
Variable-Length Semantic IDs for Recommender Systems
Feb 18, 2026Kirill KhrylchenkoSemantic ID Recommendation Generative Recommendation
HSE University, Moscow, Russia
The Best of the Two Worlds: Harmonizing Semantic and Hash IDs for Sequential Recommendation
Dec 11, 2025Ziwei Liu, Yejing Wang, Wanyu Wang +6Semantic ID Recommendation Sequential Recommendation
City University of Hong Kong Hong Kong, China · Tongji University Shanghai, China · Xi’an Jiaotong University Xi’an, China +3

May 12, 2026J/K move · Enter open · S save

Conditional Memory Enhanced Item Representation for Generative Recommendation

Ziwei Liu, Yejing Wang, Shengyu Zhou, Xinhang Li+1

City University of Hong Kong Hong Kong, China · Independent Researcher Beijing, China · Tsinghua University Beijing, China

Semantic ID Recommendation Representation Learning Recommender Systems Generative Recommendation

Open paper AskPDF arXiv

Generative recommendation (GR) has emerged as a promising paradigm that predicts target items by autoregressively generating their semantic identifiers (SID). Most GR methods follow a quantization-representation-generation pipeline, first assigning each item a SID, then constructing input representations from SID-token embeddings, and finally predicting the target SID through autoregressive generation. Existing item-level representation constructions mainly take two forms: directly merging SID-token embeddings into a compact vector, or enriching item-level representations with external inputs through additional networks. However, these item-level constructors still expose two practical challenges: direct merging may amplify the information loss caused by quantization and ID collision while obscuring SID code relations, whereas external-input-based methods can strengthen item semantics but cannot reliably preserve the SID-structured evidence required for token-level generation. These limitations make representation construction an underexplored bottleneck, leading to two severe problems, \ie{} the Identity-Structure Preservation Conflict and Input-Output Granularity Mismatch. To this end, we propose ComeIR, a Conditional Memory enhanced Item Representation framework that reconstructs SID-token embeddings into item-aware inputs and restores the token granularity during SID decoding. Specifically, MM-guided token scoring adaptively estimates the contribution of each code within the SID, dual-level Engram memory captures intra-item code composition and inter-item transition patterns, and a memory-restoring prediction head reuses the memories during SID decoding. Extensive experiments demonstrate the effectiveness and flexibility of ComeIR, and further reveal scalable gains from enlarging conditional memory.