Interview with JD Infrastructure Chief Architect He Xiaofeng on Real‑time Computing and Product Data Mining

He Xiaofeng, Chief Architect of JD Mall Infrastructure Department and speaker at the JD Technology 11.11 Infrastructure Summit, has 19 years of frontline R&D experience, having contributed to major sales events and accumulated expertise in elastic computing, middleware, and large‑scale distributed systems.

What are you currently responsible for? He oversees the computing platform and product data mining, building a real‑time computing platform on JDOS that schedules Flink, Spark, and Storm clusters, integrates storage and middleware, and supports scenarios such as product data mining, monitoring, and risk control.

He gives an example of cleaning product titles: removing irrelevant color words that merchants add for SEO, using image recognition to ensure titles match the actual product, thereby improving user experience.

He also leverages product review data to extract valuable comments, creating product impressions and tags that help users make purchase decisions and enhance search relevance.

What core technologies are used for data mining? The team relies heavily on streaming computing frameworks, machine learning, and natural language processing. Online detection accuracy exceeds 98%, and nearly 2 billion product records have been corrected. The corrected data will later be used to optimize search indexes.

What technical research will you focus on next? The plan is to strengthen the technology platform and product data mining, improve image recognition and real‑time computation capabilities, and deeply customize open‑source computing frameworks to meet large‑scale business demands.

They aim to build a product knowledge base and enable semantic reasoning—for example, understanding that a user searching for “winter coat” should also see relevant items even if the term “winter” is not explicitly in the product description—thereby improving search accuracy and user experience.

What is your view on the development of knowledge‑graph technology? He notes that using machine learning to construct knowledge graphs and perform data mining has become a major industry focus; knowledge graphs are a foundational layer, and building them well is essential for smarter, intent‑aware applications.

All PPTs, texts, and videos from the JD Technology Infrastructure Summit will be released through the “JD Technology” WeChat public account.