DREAM 技术报告

摘要

工业推荐系统通常采用级联的召回、排序和重排序流水线。尽管高效，这些流水线却将信息和目标分散到各个模块，依赖刚性规则，且对实时意图的感知能力有限，导致浏览、比较和购买之间的会话级转换未能得到充分处理。我们提出了DREAM（Developing Recommender Engine with Agentic Methods，基于智能体方法的推荐引擎开发），一种自主优化控制架构，它在不替换现有流水线的前提下，在其之上增加了一个具有感知能力、可编排且可审计的策略层。DREAM包含两个核心组件。第一，三层意图引擎（Intent Engine）将设备端信号融合为结构化的L0/L1/L2意图表示；其边缘-云端触发链将上报量压缩至约8.7%。第二，元引擎（Meta Engine）利用元模型（MetaModel）进行M1到M2再到M3的分层推理：意图摘要、由策略记忆（Strategy Memory）指导的策略规划，以及参数转换。它通过带有安全护栏的统一出口分发生成的参数。奖励双循环（Reward Dual Loop）通过将离线模拟用于策略空间探索与在线反馈用于结果校准相结合，持续优化两个组件，形成生成、执行、评估和经验积累的闭环。在淘宝首页信息流上的大规模A/B测试表明，仅对重排序进行控制即可使IPV提升2.06%，核心IPV提升2.39%，GMV提升0.88%。将控制扩展至精排后，这些增益分别提升至2.71%、3.06%和1.31%，同时PV持续提升超过1%。这些增益既无需替换流水线模型，也不牺牲服务稳定性，支持智能体元控制作为工业推荐的可行范式。

English

Industrial recommender systems commonly use cascaded retrieval, ranking, and re-ranking pipelines. Although efficient, these pipelines fragment information and objectives across modules, rely on rigid rules, and have limited awareness of real-time intent, leaving session-level shifts among browsing, comparison, and purchase insufficiently addressed. We present DREAM (Developing Recommender Engine with Agentic Methods), an autonomous optimization control architecture that adds a perception-aware, orchestrable, and auditable policy layer atop existing pipelines without replacing them. DREAM has two core components. First, a three-tier Intent Engine fuses on-device signals into structured L0/L1/L2 intent representations; its edge-cloud trigger chain reduces reporting volume to approximately 8.7%. Second, a Meta Engine uses a MetaModel for layered M1-to-M2-to-M3 reasoning: intent summarization, strategy planning informed by Strategy Memory, and parameter translation. It dispatches the resulting parameters through a unified outlet with safety guardrails. A Reward Dual Loop continuously optimizes both components by combining offline simulation for strategy-space exploration with online feedback for outcome calibration, forming a cycle of generation, execution, evaluation, and experience accumulation. Large-scale A/B tests on Taobao's homepage feed show that re-ranking control alone improves IPV by 2.06%, Core IPV by 2.39%, and GMV by 0.88%. Extending control to fine ranking raises these gains to 2.71%, 3.06%, and 1.31%, respectively, while consistently improving PV by more than 1%. These gains require neither replacement of pipeline models nor compromise of serving stability, supporting agentic meta-control as a viable paradigm for industrial recommendation.