How Cutting-Edge AI Models Are Revolutionizing E‑Commerce CTR Prediction

The following five research papers from JD Retail Technology were accepted at SIGIR 2025, presenting advanced AI methods for e‑commerce search and advertising.

Graph Isomorphism Network-Based Cohort Modeling in Click-Through Rate Prediction

Chinese Title: 基于图同构网络的群组建模在点击率预测中的应用

Download URL: https://dl.acm.org/doi/10.1145/3726302.3731936

Authors: Xuan Ma, Hao Peng, Jia Duan, Zhanhao Ye, Langlang Ye, Zehua Zhang, Jie He, Changping Peng, Zhangang Lin

Abstract: Click‑through rate (CTR) prediction suffers from cold‑start problems for new users. Existing encoder‑decoder approaches generate simplistic virtual behavior representations, limiting user interest expression and model generalization. This work proposes a Graph Isomorphism Network (GIN) based cohort modeling method that captures high‑order user‑item interactions, reducing embedding bias and enhancing generalization. Experiments on public and industrial datasets show significant improvements for both active and cold‑start users.

Post-event Modeling via Causal Optimal Transport for CTR Prediction

Chinese Title: 基于因果最优传输的后验信息建模用于CTR预测

Download URL: https://dl.acm.org/doi/10.1145/3726302.3731942

Authors: Yizhou Sang, Congcong Liu, Yuying Chen, Zhiwei Fang, Xue Jiang, Changping Peng, Zhangang Lin, Ching Law, Jingping Shao

Abstract: Accurate CTR prediction relies on post‑event features that are unavailable during inference, causing training‑inference mismatch and low coverage. The proposed Causal Optimal Transport (COT) framework generates pseudo post‑event features via semi‑supervised labeling, adjusts their distribution with a Causal Distribution Shaper, and aligns feature distributions through optimal transport, improving user interest modeling and bias mitigation. Experiments on real data demonstrate superior performance and theoretical analysis confirms robustness.

ADORE: Autonomous Domain‑Oriented Relevance Engine for E‑commerce

Chinese Title: 基于领域自适应的电商相关性判别系统

Download URL: https://dl.acm.org/doi/10.1145/3726302.3731944

Authors: Ming Pang, Chunyuan Yuan, Xiaoyu He, Zheng Fang, Donghao Xie, Fanyi Qu, Xue Jiang, Changping Peng, Zhangang Lin, Zheng Luo, Jingping Shao

Abstract: To address data scarcity and weak inference of shallow online models in e‑commerce relevance tasks, this work introduces a chain‑of‑thought reasoning large model that automatically generates domain‑specific hard samples and aligns them with online user behavior via a KTO reinforcement learning algorithm. An error‑type‑aware generation model creates adversarial samples for difficult scenarios. Knowledge is transferred to shallow models using key attribute features from COT analysis, yielding significant gains in relevance and advertising revenue in large‑scale online A/B tests.

Multi‑objective Aligned Bidword Generation Model for E‑commerce Search Advertising

Chinese Title: 多目标对齐广告买词生成模型用于电商搜索广告

Download URL: https://arxiv.org/abs/2506.03827

Authors: Zhenhui Liu, Chunyuan Yuan, Ming Pang, Zheng Fang, Li Yuan, Xue Jiang, Changping Peng, Zhangang Lin, Zheng Luo, Jingping Shao

Abstract: The paper proposes MoBGM, a multi‑objective aligned bidword generation model comprising a discriminator, generator, and preference‑alignment module. The discriminator predicts relevance, authenticity, and revenue scores for query‑rewrite pairs; the generator is trained to maximize the combined objective. Extensive offline and online experiments show MoBGM outperforms state‑of‑the‑art methods, delivering substantial commercial value.

Hierarchical User Long‑term Behavior Modeling for Click‑Through Rate Prediction

Chinese Title: 层次化用户长期行为建模在点击率预估中的应用

Download URL: https://dl.acm.org/doi/10.1145/3726302.3730207

Authors: Mao Pan, Xuanhua Yang, Nan Qiao, Dongyue Wang, Feng Mei, Xiwei Zhao, Sulong Xu

Abstract: Industrial CTR prediction relies on Transformer‑based networks, but long user behavior sequences challenge inference latency. The proposed Hierarchical Behavior Modeling (HBM) network first routes long‑term behavior into multiple interest clusters, then selects top‑k interests via a fine‑grained interest network, and finally applies a Transformer to model sequences related to these interests while capturing inter‑interest relations. Online A/B tests on JD’s platform show large performance gains.