Generative recommendation improves item predictions with balanced semantic codes
Structure-Aligned History Learning with Subspace-Structured Semantic IDs for Generative Recommendation
Information Retrieval
Summary
Making good recommendations means understanding what users like by looking at their past choices. The paper shows a better way to break down item information into balanced pieces and use them to learn from a user’s entire history, not just the latest choice. This helps the system capture long-lasting preferences more accurately. The authors tested their approach on Amazon Reviews and saw clear improvements over past methods.
What this means in practice
- •For ecommerce recommendation teams: Improve personalized product suggestions by encoding item histories with balanced, structure-aligned semantic IDs to capture user preferences over time.
- •For streaming service developers: Enhance viewing or listening recommendations by reconstructing user interaction history using balanced tokens and subspace structures for better next-item predictions.
Authors
Kaiwen Wei, Kejun He, Jie Zhang, Xiaomian Kang, Yuming Yang, Aijia Zhou, Xie Zhou, Yining Wang, Nayu Liu, Li Jin, Zhenyang Li, Jiang Zhong, He Bai, Jingxuan Wei, Junnan Zhu
Abstract
Generative recommendation constructs semantic IDs for catalog items and predicts the next ID from user interaction histories. However, existing methods face limitations in both stages. First, conventional quantizers may produce imbalanced code assignments and lack coherent intermediate semantic units between individual tokens and complete items, limiting history reconstruction. Second, recommendation models remain dominated by next-item prediction, treating historical interactions mainly as conditioning context and providing limited supervision for recovering persistent preferences across the historical path, particularly when relevant preference signals are distributed over distant interactions. To address these limitations, Structure-Aligned History Learning (SAHL) reconnects the two stages by using the structure created during quantization directly for history reconstruction. Its Semantic Residual Balanced Quantization (SRBQ) forms a balanced token-subspace-item hierarchy, and SAHL reconstructs history at these same granularities. Predictive entropy prioritizes uncertain transitions, while curriculum optimization progressively shifts training to inference-aligned next-item prediction. On four Amazon Review categories, SAHL improves NDCG@10 by 18.6%-30.0% over RPG and 2.7%-6.0% over strong method MHL. Controlled analyses verify balanced code use, the benefit of following quantizer-defined groups during reconstruction, and dependence on user history. The relevant code is available at https://github.com/CQU-MM-Intelligent-Lab/SAHL.