Tech Insight: Highlights of Ten JD Retail Technology Papers Published in Top AI Conferences (2024)

Tech Insight, a JD retail technology column, introduces ten recent research papers accepted at leading AI conferences (CVPR, SIGIR, WWW, AAAI, IJCAI) in 2024, spanning object detection, ranking, recommendation, dataset construction, query classification, click‑through‑rate prediction, and large‑language‑model hallucination mitigation.

01. CVPR 2024 – Exploring Region‑Word Alignment in Built‑in Detector for Open‑Vocabulary Object Detection Chinese title: 探索内置检测器中的区域‑词对齐以实现开放词汇目标检测 Authors: Heng Zhang, Qiuyu Zhao, Linyu Zheng, Hao Zeng, Zhiwei Ge, Tianhao Li, Sulong Xu Download: https://ieeexplore.ieee.org/document/10656464 Abstract: The paper proposes BIND, a framework that eliminates the need for module replacement or knowledge transfer to existing detectors by jointly training an image‑text dual encoder for fine‑grained region‑word alignment and a DETR‑style decoder for detection, together with an anchor proposal network to improve efficiency.

02. SIGIR 2024 – A Unified Search and Recommendation Framework based on Multi‑Scenario Learning for Ranking in E‑commerce Chinese title: 基于多场景学习的搜推联合建模统一框架 Authors: Jinhan Liu, Qiyu Chen, Junjie Xu, Junjie Li, Baoli Li, Sulong Xu Download: https://arxiv.org/abs/2405.10835 Abstract: The authors design a unified framework that extracts user interest and scenario‑independent features, introduces a global label space multi‑task layer, and demonstrates significant performance gains on industrial e‑commerce datasets.

03. SIGIR 2024 – Optimizing E‑commerce Search: Toward a Generalizable and Rank‑Consistent Pre‑Ranking Model Chinese title: 优化电商搜索：构建有泛化性和排序一致性的粗排模型 Authors: Enqiang Xu, Yiming Qiu, Junyang Bai, Ping Zhang, Dadong Miao, Songlin Wang, Guoyu Tang, Lin Liu, Mingming Li Download: https://arxiv.org/abs/2405.05606 Abstract: The paper proposes a pre‑ranking model that jointly optimizes ranking consistency and long‑tail generalization by adding binary classification tasks and contrastive representation learning, achieving notable offline and online improvements.

04. SIGIR 2024 – A Preference‑oriented Diversity Model Based on Mutual‑information in Re‑ranking for E‑commerce Search Chinese title: 京东搜索重排：基于互信息的用户偏好导向模型 Authors: Huimu Wang, Mingming Li, Dadong Miao, Songlin Wang, Guoyu Tang, Lin Liu, Sulong Xu, Jinghe Hu Download: https://dl.acm.org/doi/10.1145/3626772.3661359 Abstract: PODM‑MI introduces a variational Gaussian user‑preference model and maximizes mutual information between user preference and candidate items to balance relevance and diversity in re‑ranking.

05. JDivPS – A Diversified Product Search Dataset Chinese title: 基于京东电商平台的多样化产品搜索数据集 Authors: Zhirui Deng, Zhicheng Dou, Yutao Zhu, Xubo Qin, Pengchao Cheng, Jiangxu Wu, Hao Wang Download: https://dl.acm.org/doi/10.1145/3626772.3657888 Abstract: JDivPS is the first publicly available e‑commerce product search dataset with manually annotated user intents, containing 10,000 queries and ~1.68 M unique products, enabling robust evaluation of diversified search models.

06. WWW 2024 – A Semi‑supervised Multi‑channel Graph Convolutional Network for Query Classification in E‑commerce Chinese title: 基于半监督多通道图神经网络的类目预估方法 Authors: Chunyuan Yuan, Ming Pang, Zheng Fang, Xue Jiang, Changping Peng, Zhangang Lin Download: https://arxiv.org/abs/2408.01928 Abstract: SMGCN leverages similarity scores, co‑occurrence, and semantic graphs to enrich noisy posterior labels, achieving superior query intent classification performance in large‑scale e‑commerce settings.

07. WWW 2024 – PPM: A Pre‑trained Plug‑in Model for Click‑through Rate Prediction Chinese title: 基于预训练的插件式CTR预估模型 Authors: Yuanbo Gao, Peng Lin, Dongyue Wang, Feng Mei, Xiwei Zhao, Sulong Xu, Jinghe Hu Download: https://arxiv.org/abs/2403.10049 Abstract: The authors propose MoRec + URM, a two‑stage architecture that pre‑trains a multimodal encoder (MoRec) with CTR supervision and then jointly trains it with an ID‑based model (URM) for improved generalization on long‑tail items.

08. AAAI 2024 – Parallel Ranking of Ads and Creatives in Real‑Time Advertising Systems Chinese title: 京东创意优选：广告商品排序和广告创意优选的并行排序实践 Authors: Zhiguang Yang, Lu Wang, Chun Gan, Liufang Sang, Haoran Wang, Wenlong Chen, Jie He, Changping Peng, Zhangang Lin, Jingping Shao Download: https://arxiv.org/abs/2312.12750 Abstract: The paper introduces a parallel inference architecture where ad‑creative selection and product ranking run simultaneously, sharing latency budget and delivering higher relevance and diversity without extra online cost.

09. AAAI 2024 – Generalize for Future: Slow and Fast Trajectory Learning for CTR Prediction Chinese title: 面向未来的泛化：用于点击率预测的慢速和快速轨迹学习 Authors: Jian Zhu, Congcong Liu, Xue Jiang, Changping Peng, Zhangang Lin, Jingping Shao Download: https://ojs.aaai.org/index.php/AAAI/article/view/27797 Abstract: The authors propose a three‑learner framework (fast, slow, and working learners) with a trajectory loss to mitigate domain drift and improve temporal adaptation in CTR models.

10. IJCAI 2024 – TaD: A Plug‑and‑Play Task‑Aware Decoding Method to Better Adapt LLMs on Downstream Tasks Chinese title: TaD+RAG‑缓解大模型“幻觉”的组合新疗法 Authors: Xinhao Xu, Hui Chen, Zijia Lin, Jungong Han, Lixing Gong, Guoxin Wang, Yongjun Bao, Guiguang Ding Download: https://www.ijcai.org/proceedings/2024/728 Abstract: TaD introduces a task‑aware decoding strategy that compares supervised fine‑tuned outputs to mitigate hallucinations in LLMs; combined with Retrieval‑Augmented Generation (RAG), it offers a generic, plug‑and‑play solution for diverse downstream tasks.