JD's AI-Driven Image Automation and Knowledge Graph Applications in E-commerce

In e-commerce promotion and sales, many advertisements rely on images, which were previously edited by professional designers, limiting processing speed. Since early 2018, JD has widely adopted automated image synthesis technology internally.

To achieve this goal, the first step is machine intelligence cutout, enabling full-category product subject recognition, precise localization and segmentation, and batch intelligent cutout. The second step involves using machines to learn the aesthetic design principles of user experience designers to achieve intelligent layout. The third step is to realize high-performance batch image synthesis.

Liu Haifeng summarized that machine-generated images currently account for about 1% of total product volume, with the proportion expected to rise to 15%-20% in two years, meaning most promotional product images will be machine-generated. JD also performs automated image translation to support its globalization strategy of selling domestic products abroad, combining intelligent layout technology to achieve rapid global product distribution.

Based on JD's massive product data, the company has conducted knowledge graph work—mining product tag relationships—and deeply explored various data attributes to construct a large-scale graph model knowledge base, which serves as a foundation to empower many intelligent applications.

For example, the JIMI intelligent customer service robot can provide real-time, intelligent, fast responses to common user inquiries, delivering personalized, one‑to‑one answers for different users, and it can also show empathy by proactively caring for users based on their profiles.

Additionally, there is the 'Li Bai' writing robot, a deep‑learning‑based natural language technology product tailored for e‑commerce content creation, aimed at improving writing efficiency and quality. It can automatically generate marketing copy, focus‑point image descriptions, and other short texts for specific products; it is already used at scale in JD's 'Discover Good Products' channel, and many third‑party merchants also employ Li Bai for promotional copywriting.

Liu Haifeng noted that in the big‑data application domain for massive product catalogs, JD leverages unified storage, elastic computing, visual intelligence, and knowledge graphs to support a range of product applications, and will continue investing in this area to enhance user experience.