Congcong Liu
Senior Algorithm Expert at JD Retail and Head of Search Advertising Ranking Models at JD.com
He received his Bachelor’s degree in Engineering from Zhejiang University and his Ph.D. from the Hong Kong University of Science and Technology through direct admission. He was selected for the Beijing Overseas High-Level Talent Introduction Program. In 2021, he joined JD.com, where he has been responsible for the deployment and innovation of JD’s search advertising ranking models. In recent years, he has led several major projects to full-scale online deployment and iterative upgrades, including concept drift adaptation, post-event distillation, cross-domain loss harmonizing, and large-scale synchronous training optimization for ranking large models. These projects have resulted in more than 10 papers published at top AI conferences such as WWW, KDD, SIGIR, AAAI, RecSys, and CIKM (including 6 as first author), as well as over 10 patents (including 6 as first inventor).
Topic
Application of Generative Large Models in Advertising Ranking
Advertising ranking is a core business for e-commerce platforms and faces complex real-world scenarios. Ad systems must accurately understand users’ rapidly changing intents and vast dynamic product information, creating a challenging semantic matching and sequence modeling problem. Platforms need to leverage advanced model capabilities to simultaneously improve click-through rates, optimize user experience, and increase exposure for long-tail products. This is particularly challenging due to ambiguous user intents, sparse behavior sequences, and cold-start issues for new ads, all of which make precise prediction difficult. Our work focuses on applying generative large models within ranking models. By using sequence generation-based modeling, we achieve a unified understanding of both users and ad content, significantly enhancing the model’s generalization and adaptability to dynamic scenarios and unseen cases, resulting in substantial improvements in business performance.
