Building a Data Middle Platform: Practices and Architecture at NetEase Yanxuan

Data middle platforms, first mentioned by Alibaba, gained widespread attention around 2018 as many companies began to adopt them for large‑scale data applications. Enterprises increasingly seek data‑driven transformation and fine‑grained operations, prompting the construction of such platforms despite their significant investment.

NetEase Yanxuan started planning its data middle platform in 2017 when data volume grew sufficiently. The platform now supports search, recommendation, BI reporting, data dashboards, CRM, DMP, and risk control, all of which rely on robust underlying big‑data capabilities.

The data middle platform is defined as a combination of high‑quality, efficient data systems and services that empower the data front‑end. Its core responsibility is to enable the front‑end (search, recommendation, BI, etc.) to deliver business value efficiently.

Architecturally, the platform consists of three layers: the data warehouse layer (handling storage and computation), the data services layer (providing unified query, tagging, metric monitoring, and data output services), and the BI platform layer (supporting agile reporting and visualization).

The data warehouse layer includes the core warehouse and management systems such as the tracking management system (Kua Fu) and data entry system (Jing Wei) to ensure data completeness and quality, as well as metric management (Cang Jie) and metric mapping (Sui Ren) for consistency.

The data services layer centralizes access through a unified query service and a tagging service, acting as a gateway between the warehouse and downstream applications, with features like model‑level rate limiting and circuit breaking.

The BI platform, Yanxuan YouShu, is an agile BI tool that offers PPT‑like operations, high flexibility, and performance optimizations via caching strategies, achieving near‑100% first‑visit cache hit rates.

In summary, when an enterprise requires data‑driven transformation and fine‑grained operational capabilities that generate large‑scale data application demands, building a data middle platform becomes essential. Such a platform combines a data warehouse system, a suite of data services, and a BI platform to efficiently empower the data front‑end.