Structuring and Tokenizing Distributed User Interest Context for Generative Recommendation

论文概要

研究领域: ML 作者: Ruizhong Qiu, Yinglong Xia, Dongqi Fu 发布时间: 2025-06-23 arXiv: 2506.18494

中文摘要

生成式推荐是工业推荐系统中一种新兴范式，旨在从用户历史行为预测其下一次交互。生成式推荐的核心在于物品token化，它连接物品语义和推荐模型。然而，现有方法往往难以同时有效地组织和注入复杂的用户行为上下文和物品语义上下文到推荐模型中。一方面，现有基于图的集成方法，如图序列化和图神经网络，要么存在可扩展性问题，要么仅利用局部图信息；另一方面，现有语义token化方法通常依赖启发式方法且缺乏显式监督信号，可能导致不准确或次优的语义表征。为解决用户兴趣上下文建模中的这些局限性，本文提出G2Rec，一个可扩展框架，统一整体图用户共参与建模与语义token化，用于工业级生成式推荐。总体而言，G2Rec使推荐模型能够在不需要真实用户兴趣的情况下捕获整体且语义基础的用户兴趣原型，从而提供更全面和准确的用户行为上下文建模。在线部署和大量实验证明了G2Rec相对于现有方法的优越性。

原文摘要

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, whi...

--- *自动采集于 2026-06-23*

#论文 #arXiv #ML #小凯