深度拆解One4All框架,生成式推荐在京东CPS广告的技术探索
背景介绍
现有方案汇总: 落地页商品推荐是站外广告很重要的一种形式,对应的研究课题是触发诱导推荐 (Trigger-Induced Recommendation, TIR),现有方案包含如下三类: •基于历史行为序列隐式建模用户意图; •利用触发项进行I2I召回或通过sku2query生成搜索词再进行商品检索; •通过三个网络来分别表示触发项、建模用户历史行为和预估权重来融合前两者,例如,DIHN、DIAN、DEI2N和DUIN[8、9、10、11]。解决方案
(1) 通过传统推荐数据自动化地生成丰富的意图描述,以意图文本+历史商品语义ID序列作为输入,目标商品语义ID作为输出的方式 (2) 重构触发诱导推荐的任务范式,利用 (3) 生成式指令遵循微调的方式实现对历史行为和触发项的感知和动态调控。 自动化意图生成和评估 •输入“用户行为数据+目标商品”; •基于Few-shot Prompting和CoT策略,通过言犀-81B模型对用户行为数据进行总结、推理,并预测当前意图; •输出“总结-推理-预测”的三元组数据; •利用Self-Verification的方式对生成的显性意图进行评估。线上模型更新策略 方案效果 •完成了线上例行化的开发,支撑CPS广告每天1000W+UV的在线实时推理; •基于One4All生成式推荐框架,在现有序列推荐主任务的基础上兼容更多的行为和语言理解的任务,推动召排一体化、搜推联合建模、用户行为总结、个性化意图推断等技术的探索。 06 总结和未来展望 •交互式推荐系统(搜索推荐联合) ◦现有方案仍未更大限度激发生成式模型的效果和能力,交互式应用是值得尝试的方向,同时需要配合上下游进行产品形式的重构。 •多模态信息理解与生成 ◦前链路中有丰富的图片和视频信息,对多模态信息进行高效地理解和内容组织,可以增强推荐效果和提升展示形式的丰富程度。 07 参考文献 1.Xu L, Zhang J, Li B, et al. Prompting large language models for recommender systems: A comprehensive framework and empirical analysis[J]. arXiv preprint arXiv:2401.04997, 2024. 2.知乎《一文梳理工业界大模型推荐实战经验》. 2024 3.Zhai J, Liao L, Liu X, et al. Actions speak louder than words: trillion-parameter sequential transducers for generative recommendations[C]//Proceedings of the 41st International Conference on Machine Learning. 2024: 58484-58509. 4.Chen J, Chi L, Peng B, et al. Hllm: Enhancing sequential recommendations via hierarchical large language models for item and user modeling[J]. arXiv preprint arXiv:2409.12740, 2024. 5.Zhang C, Wu S, Zhang H, et al. Notellm: A retrievable large language model for note recommendation[C]//Companion Proceedings of the ACM Web Conference 2024. 2024: 170-179. 6.Zhang C, Zhang H, Wu S, et al. NoteLLM-2: multimodal large representation models for recommendation[J]. arXiv preprint arXiv:2405.16789, 2024. 7.Deng J, Wang S, Cai K, et al. OneRec: Unifying Retrieve and Rank with Generative Recommender and Iterative Preference Alignment[J]. arXiv preprint arXiv:2502.18965, 2025. 8.Ma J, Xiao Z, Yang L, et al. Modeling User Intent Beyond Trigger: Incorporating Uncertainty for Trigger-Induced Recommendation[C]//Proceedings of the 33rd ACM International Conference on Information and Knowledge Management. 2024: 4743-4751. 9.Shen Q, Wen H, Tao W, et al. Deep interest highlight network for click-through rate prediction in trigger-induced recommendation[C]//Proceedings of the ACM web conference 2022. 2022: 422-430. 10.Xia Y, Cao Y, Hu S, et al. Deep intention-aware network for click-through rate prediction[C]//Companion Proceedings of the ACM Web Conference 2023. 2023: 533-537. 11.Xiao Z, Yang L, Zhang T, et al. Deep evolutional instant interest network for ctr prediction in trigger-induced recommendation[C]//Proceedings of the 17th ACM International Conference on Web Search and Data Mining. 2024: 846-854. 12.Ma J, Zhao Z, Yi X, et al. Modeling task relationships in multi-task learning with multi-gate mixture-of-experts[C]//Proceedings of the 24th ACM SIGKDD international conference on knowledge discovery & data mining. 2018: 1930-1939. 13.Tang H, Liu J, Zhao M, et al. Progressive layered extraction (ple): A novel multi-task learning (mtl) model for personalized recommendations[C]//Proceedings of the 14th ACM conference on recommender systems. 2020: 269-278. 14.Ma X, Zhao L, Huang G, et al. Entire space multi-task model: An effective approach for estimating post-click conversion rate[C]//The 41st International ACM SIGIR Conference on Research & Development in Information Retrieval. 2018: 1137-1140. 15.Zhou Z, Liu J, Shao J, et al. Beyond One-Preference-Fits-All Alignment: Multi-Objective Direct Preference Optimization[C]//Findings of the Association for Computational Linguistics ACL 2024. 2024: 10586-10613. 16.Li K, Zhang T, Wang R. Deep reinforcement learning for multi-objective optimization[J]. IEEE transactions on cybernetics, 2020, 51(6): 3103-3114. 17.Rame A, Couairon G, Dancette C, et al. Rewarded soups: towards pareto-optimal alignment by interpolating weights fine-tuned on diverse rewards[J]. Advances in Neural Information Processing Systems, 2023, 36: 71095-71134. 18.Rafailov R, Sharma A, Mitchell E, et al. Direct preference optimization: Your language model is secretly a reward model[J]. Advances in Neural Information Processing Systems, 2023, 36: 53728-53741. 19.Wu J, Xie Y, Yang Z, et al. beta-DPO: Direct Preference Optimization with Dynamic beta[J]. Advances in Neural Information Processing Systems, 2025, 37: 129944-129966. 20.Lin X, Chen H, Pei C, et al. A pareto-efficient algorithm for multiple objective optimization in e-commerce recommendation[C]//Proceedings of the 13th ACM Conference on recommender systems. 2019: 20-28. 21.Hu J, Tao L, Yang J, et al. Aligning language models with offline learning from human feedback[J]. arXiv preprint arXiv:2308.12050, 2023. 22.Yang R, Pan X, Luo F, et al. Rewards-in-context: multi-objective alignment of foundation models with dynamic preference adjustment[C]//Proceedings of the 41st International Conference on Machine Learning. 2024: 56276-56297. 推荐阅读👇 多智能体强化学习的算力调度创新,让每一份算力都创造广告价值 | 京东零售技术实践 WWW2025 论文解读 | 京东零售广告创意:引入场域目标的创意图片生成Model_DPO_T和Model_SFT_T进行比较,选出优胜模型Aif cvr+且ctr降低小于10%:Model_A = Model_DPO_Telse:Model_A = Model_SFT_TModel_BEST_T-1和Model_A进行比较,选出优胜模型Model_BEST_T上线if cvr+且ctr降低小于10%:Model_BEST_T = Model_Aelse:Model_BEST_T = Model_BEST_T-1
基于DPO多目标优化:京东生成式推荐的探索与实践 | 京东零售技术实践