Modern feed recommendation and search systems are deeply connected in user behavior but are usually modeled by separate architectures. Feed recommendation mainly captures implicit interests from browsing interactions, while search systems rely on explicit user queries to retrieve intent-matched content. This separation causes fragmented user understanding and missed opportunities for using feed interactions to improve query generation and using generated queries to enhance feed candidate retrieval. In this paper, we propose OneFeed, a unified generative framework for jointly modeling feed content enhancement and query generation. OneFeed encodes heterogeneous user behavior sequences with a shared behavior encoder and employs two generative heads: a Feed Semantic ID Generator that produces content semantic IDs for recommendation retrieval, and an Intent Query Generator that produces natural-language queries for search-based candidate expansion. To bridge the semantic gap between recommendation content and search queries, we introduce a SID-Query alignment objective that learns a shared semantic space for content semantic IDs and query representations. We further design a closed-loop self-enhancement paradigm that leverages implicit user feedback from generated content and search-retrieved results to improve both generation tasks. We report measured offline replay results on public datasets (MovieLens-1M and Amazon Reviews) under a torch-free prototype, alongside a detailed experimental protocol, a comprehensive set of evaluation metrics, and an analysis of where the unified framework helps and where gains await learned semantic IDs and query generators. OneFeed provides a practical and extensible direction for unifying search and recommendation through generative modeling.

翻译：现代推送推荐系统与搜索系统在用户行为层面紧密关联，却通常采用独立架构进行建模。推送推荐主要通过浏览交互捕捉隐式兴趣，而搜索系统依赖用户显式查询以检索匹配意图的内容。这种割裂导致用户理解碎片化，既错失利用推送交互改善查询生成的机会，也无法通过生成查询增强推送候选内容的检索能力。本文提出统一化生成式框架OneFeed，实现对推送内容增强与查询生成的联合建模。该框架采用共享行为编码器处理异构用户行为序列，并配备两个生成模块：用于推荐检索的内容语义ID生成器，以及用于基于搜索的候选扩展的自然语言意图查询生成器。为弥合推荐内容与搜索查询之间的语义鸿沟，我们引入语义ID-查询对齐目标，在内容语义ID与查询表征之间建立共享语义空间。进一步设计闭环自增强范式，利用生成内容与搜索结果中的隐式用户反馈，同时优化两项生成任务。我们在无torch原型条件下，于公开数据集（MovieLens-1M与Amazon Reviews）上实测离线回放结果，并附详细实验方案、综合评估指标体系，剖析统一框架的优势所在及待改进方向（包括优化学习型语义ID与查询生成器）。OneFeed为通过生成式建模统一搜索与推荐系统提供了兼具实用性与扩展性的研究方向。