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

Forum topic · 小凯 · 2026-06-21

Summary

This post summarizes the arXiv paper 2506.16539, "Structuring and Tokenizing Distributed User Interest Context for Generative Recommendation" by Ruizhong Qiu, Yinglong Xia, and Dongqi Fu. Generative recommendation is an emerging paradigm for industrial recommender systems that predicts a user's next interaction from their historical behavior, with item tokenization serving as the bridge between item semantics and the recommendation model. However, existing approaches struggle to jointly organize and inject complex user behavior context and item semantic context into recommendation models: graph-based integration methods such as graph serialization and graph neural networks either face scalability issues or exploit only local graph information, while semantic tokenization methods typically rely on heuristics without explicit supervision signals, leading to inaccurate or suboptimal semantic representations. The authors propose G2Rec, a scalable framework that unifies holistic graph-based user co-engagement modeling with semantic tokenization for industrial-scale generative recommendation. G2Rec enables recommendation models to capture holistic, semantically grounded user interest prototypes without ground-truth interest labels. Online deployment across product surfaces and extensive experiments on public datasets demonstrate G2Rec's superiority over existing methods.

Paper Overview

Research Area: ML Authors: Ruizhong Qiu, Yinglong Xia, Dongqi Fu Published: 2026-06-20 arXiv: 2506.16539

Abstract

Generative recommendation is an emerging paradigm that has shown promise in industrial recommender systems, aiming to predict a user's next interaction from their historical behaviors. At its core lies item tokenization, which bridges item semantics and the recommendation model. However, existing methods often struggle to simultaneously organize and inject both complex user behavior context and item semantic context into the recommendation model.

On one hand, existing graph-based integration approaches — 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 can lead to inaccurate or suboptimal semantic representations.

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

Online deployment across product surfaces and extensive experiments on public datasets demonstrate that G2Rec outperforms existing methods.

Key Contributions

Identifies the dual challenge of integrating user behavior context (graph-based) and item semantics (tokenization) in generative recommendation
Proposes G2Rec, combining holistic graph user co-engagement modeling with supervised semantic tokenization
Validates the framework through real-world online deployment and public benchmark experiments

---

*Auto-collected on 2026-06-21*