Smart DMP: A Next‑Generation Intelligent Targeting System for E‑commerce Advertising

Targeting is a crucial component of e‑commerce marketing; traditional DMP targeting based on tag profiles and newer AI‑driven intelligent targeting each suffer from distinct drawbacks such as low tag quality control and limited merchant expressiveness.

Smart DMP, developed by JD.com, aims to resolve these issues by integrating merchant intent expression with a high‑relevance modeling layer, achieving more than a 15% increase in click‑through rate (CTR) and gaining adoption by over 50% of merchants.

The system architecture introduces a delayed‑intervention mechanism, a relevance model (RMM) that evaluates product‑user relevance using a long‑short‑term behavior model (LSBM) with self‑attention and DNN layers, and a tag‑mining model (TMM) that enriches user profiles by reusing the existing DMP tag model.

During online serving, RMM first recalls a high‑relevance ad set for the selected product, TMM generates tags for the incoming user request, the intersection of merchant‑chosen tags and user tags is retained, and the resulting ads are passed to a ranking model for top‑k recommendation.

Offline A/B experiments demonstrate substantial improvements in core metrics such as CTR, CPC, eCPM, and overall advertising efficiency compared with both traditional DMP and pure intelligent targeting.