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 showing promise in industrial recommender systems, aiming to predict a user's next interaction from their historical behaviors. At its core lies item tokenization, which connects item semantics with the recommendation model. However, existing methods often struggle to simultaneously and effectively organize and inject both complex user behavior context and item semantic context into recommendation models.
On one hand, existing graph-based integration methods (such as graph serialization and graph neural networks) either face scalability issues or only leverage local graph information. On the other hand, existing semantic tokenization approaches 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, this paper proposes G2Rec, a scalable framework that unifies holistic graph-based user co-participation 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, providing more comprehensive and accurate user behavior context modeling in industrial sequential recommendation. Online deployment across product surfaces and extensive experiments on public datasets demonstrate G2Rec's superiority over existing methods.
--- *Auto-collected on 2026-06-21*