Selected Meituan Papers Accepted at KDD 2024: Summaries of Five Long Papers

KDD (ACM SIGKDD) is the premier international conference on Knowledge Discovery and Data Mining. This article highlights five long papers from the Meituan technology team that were accepted to KDD 2024, covering topics such as user intent perception, machine learning & optimization, online controlled experiments, joint advertising models, and real‑time order pooling.

1. Unified Dual‑Intent Translation for Joint Modeling of Search and Recommendation Authors: Yuting Zhang*, Yiqing Wu*, Ruidong Han, Ying Sun, Yongchun Zhu, Xiang Li, Wei Lin (affiliations: ICT, CAS; Meituan; HKUST). Type: Long Paper. PDF . Abstract: The paper proposes UDITSR, a model that jointly captures stable (intrinsic) and dynamic (need) user intents by leveraging search queries as supervision for recommendation. A dual‑intent conversion propagation mechanism learns interpretable relationships among intrinsic intent, need intent, and interacted items. Experiments show UDITSR outperforms SOTA baselines on both search and recommendation tasks, and online A/B tests on Meituan’s food‑delivery platform report +1.46% GMV and +0.77% CTR.

2. Joint Auction in the Online Advertising Market Authors: Zhen Zhang, Weian Li, Yahui Lei, Bingzhe Wang, Zhicheng Zhang, Qi Qi (RUC, Gaoling School of AI); Qiang Liu, Xingxing Wang (Meituan). Type: Long Paper. Abstract: The paper introduces a joint auction framework that allows both brand advertisers and shop owners to bid for ad slots simultaneously. To design optimal mechanisms under this new setting, the authors propose JRegNet, a neural architecture that approximates DSIC and IR while achieving superior performance on simulated and real data.

3. STATE: A Robust ATE Estimator of Heavy‑Tailed Metrics for Variance Reduction in Online Controlled Experiments Authors: Hao Zhou*, Kun Sun* (Meituan), Shaoming Li, Yangfeng Fan, Guibin Jiang (Meituan), Jiaqi Zheng (Nanjing University), Tao Li (Meituan). Type: Long Paper. PDF . Abstract: STATE combines t‑distribution modeling with variational EM to robustly estimate average treatment effects (ATE) for heavy‑tailed business metrics. It reduces variance by ~50% compared with CUPAC/MLRATE, enabling the same statistical power with half the samples or experiment duration.

4. Decision Focused Causal Learning for Direct Counterfactual Marketing Optimization Authors: Hao Zhou* (Meituan), Rongxiao Huang* (Nanjing University), Shaoming Li, Guibin Jiang, Jiaqi Zheng, Bing Cheng, Wei Lin (Meituan). Type: Long Paper. PDF . Abstract: The paper presents DFCL, an end‑to‑end decision‑oriented causal learning framework that aligns machine‑learning predictions with downstream operations‑research optimization objectives for marketing budget allocation. DFCL handles budget uncertainty, counterfactual inference, and scalability, and has been deployed in multiple Meituan marketing scenarios.

5. Harvesting Efficient On‑Demand Order Pooling from Skilled Couriers: Enhancing Graph Representation Learning for Refining Real‑time Many‑to‑One Assignments Authors: Yile Liang, Jiuxia Zhao, Donghui Li (Meituan); Jie Feng, Chen Zhang (Tsinghua University); Xuetao Ding, Jinghua Hao, Renqing He (Meituan). Type: Long Paper. PDF . Abstract: The work builds an efficiency‑perception network that extracts order‑pooling potential from experienced couriers’ trajectories. By embedding orders into low‑dimensional vectors and performing similarity‑based pruning, the system rapidly identifies high‑quality order groups for real‑time dispatch, improving assignment quality and courier experience.

These papers demonstrate Meituan’s collaboration with academia and its contributions to AI‑driven solutions in recommendation, advertising, experimentation, causal inference, and logistics.