Xianyu Premium Product Library: Architecture and Implementation

Xianyu is a C2C idle‑goods platform where each user can be both buyer and seller. With rapid growth, product quality varies, but rich attribute data has been accumulated, prompting the need to analyze these attributes to build a premium‑product library.

The core challenge is to leverage structured product and user attributes to identify and define high‑quality items, then tag them for inclusion in the library.

The solution abstracts metric models from various dimensions (completeness, price, supply‑demand, seller) based on structured information, enabling systematic product tagging.

Instead of black‑box deep‑learning models, an interpretable knowledge‑base approach is used. Index models for each dimension are built, validated through simulation and regression analysis, and combined to score product quality.

Feature calculation involves both real‑time and offline pipelines. Real‑time updates trigger computations on the TPP platform, with a unified meta‑data layer and message bus (Metaq). Offline processing recomputes the full catalog daily on ODPS (MaxCompute). Full‑incremental data overlap is resolved using iGraph’s backup‑restore and incremental replay mechanism.

Service capabilities include multi‑dimensional online queries via HSF, real‑time change message subscription, and offline data services at day, hour, and minute granularity.

Business results show significant improvements: click‑through rates up by >20%, conversion rates nearly doubled for new‑user homepages, and overall transaction metrics increased.

Future work will explore product clustering for deeper price insights and integrate with downstream recycling flows to enhance platform value.