Data‑Driven Seller Activity Enhancement on Xianyu

Introduction: Xianyu is a C2C platform; increasing seller activity benefits transactions and user growth. The key is transaction efficiency, especially for individual sellers.

Data analysis revealed patterns: sellers with real avatars attract more visits; responsive sellers close deals faster; proximity and rich product info improve sales.

Seller behavior loop: activity is driven by transaction conversion. Based on observations, attempts were made to influence behavior.

Two key indicators for simulation: online status (current and time since last online) and inquiry reply statistics (responses in the past half hour).

The engineering solution defines four behavior elements: when, where, what, who. Complex event processing (CEP) is used to match behavior patterns. After comparison, Siddhi was chosen as the CEP engine.

Complete behavior loop: guide sellers → seller actions → generate revenue → perceive revenue. Three modules were designed: activities, tasks, and data collection.

Activities guide sellers, provide incentives, and manage metadata, queries, synchronization, and disaster recovery.

Tasks define desired seller actions, manage metadata, callbacks, initialization from algorithm output, data analysis, and buyer feedback, and ensure data consistency via optimistic‑locking‑style updates.

Data synchronization channel aggregates task‑level data to activity level, supporting incremental real‑time sync and full‑batch sync with distributed task distribution and idempotent operations.

Data layer converges seller behavior data from the standardized Omega system and heterogeneous sources (logs, MQ, storage). Depending on complexity, real‑time stream computation, CEP engine, or persistent storage are used.

Additional modules include revenue perception, audience targeting, and a service layer for protocol and template rendering.

Results: behavior‑based traffic allocation (online status & reply metrics) increased conversion by 3 percentage points; new scenarios can be launched without developer involvement.

Future plans focus on richer behavior data, personalized guidance, and enhanced feedback mechanisms.