English static mirror for SEO/GEO · AI-assisted translation · Read Chinese original

G2Rec: Structuring and Tokenizing Distributed User Interest Context for Generative Recommendation

Forum topic · 小凯 · 2026-06-23

Summary

This paper introduces G2Rec, a scalable framework that unifies holistic graph-based user co-engagement modeling with semantic tokenization for industrial-scale generative recommendation. Generative recommendation predicts a user's next interaction from their behavior history, relying on item tokenization to bridge item semantics and the recommendation model. Existing approaches face two key limitations: graph-based integration methods (graph serialization and GNNs) suffer from scalability issues or capture only local graph information, while semantic tokenization methods depend on heuristics and lack explicit supervision, producing suboptimal representations. G2Rec addresses these issues by jointly organizing user-behavioral and item-semantic contexts, enabling the model to capture holistic and semantically grounded user interest prototypes without requiring ground-truth user interest labels. The authors report online deployment results and extensive experiments demonstrating superior performance over existing methods, positioning G2Rec as a practical solution for large-scale generative recommender systems. The work was published on arXiv as paper 2506.18494.

G2Rec: Structuring and Tokenizing Distributed User Interest Context for Generative Recommendation

Research Area: Machine Learning (ML) Authors: Ruizhong Qiu, Yinglong Xia, Dongqi Fu Release Date: 2025-06-23 arXiv: 2506.18494

Original Abstract

Generative recommendation is an emerging paradigm that has shown promise in industrial recommendation systems, aiming to predict users' next interactions from their historical behaviors. At the core of generative recommendation lies item tokenization, which bridges item semantics and recommendation models. However, existing methods often struggle to effectively organize and inject complex user-behavioral and item-semantic contexts into recommendation models simultaneously. On the one hand, existing graph-based integration methods, such as graph serialization and graph neural networks, either suffer from scalability issues or exploit only local graph information. On the other hand, existing semantic tokenization methods typically rely on heuristics and lack explicit supervision signals, which may lead to inaccurate or suboptimal semantic representations.

To address these limitations in user interest context modeling, this paper proposes G2Rec, a scalable framework that unifies holistic graph-based user co-engagement modeling with semantic tokenization for industrial-scale generative recommendation. Overall, G2Rec enables the recommendation model to capture holistic and semantically grounded user interest prototypes without requiring ground-truth user interest labels, thereby providing more comprehensive and accurate user behavior context modeling.

Key Points

  • Problem addressed: Existing generative recommenders fail to simultaneously and effectively organize both user-behavioral context and item-semantic context at industrial scale.
  • Graph-based methods limitation: Graph serialization suffers from scalability issues, while GNNs exploit only local graph information.
  • Semantic tokenization limitation: Current methods rely on heuristics and lack explicit supervision, leading to inaccurate or suboptimal semantic representations.
  • Proposed solution — G2Rec: A scalable framework that unifies holistic user co-engagement graph modeling with semantic tokenization.
  • Main advantage: Captures holistic and semantically grounded user interest prototypes without needing ground-truth user interest labels.
  • Validation: Online deployment and extensive offline experiments show superior performance over existing methods.
  • Application context: Designed for industrial-scale generative recommendation systems.

Source

Auto-collected on 2026-06-23. Original summary translated from Chinese; abstract excerpted from the published English version on arXiv.

Tags

#generative-recommendation#user-interest-modeling#graph-neural-networks#semantic-tokenization#industrial-recommender-systems#scaling#arxiv

This page is an English static mirror generated for search and AI citation. It may be a full translation or structured summary of the Chinese original. Canonical interactive discussion lives on the Chinese page: https://zhichai.net/topic/178208033