Device Fingerprinting and User Growth Architecture in Alibaba's Xianyu Platform

Business background: User growth involves acquisition, activation, retention, monetization, and referral, not just raw numbers.

Problem definition: Need a method to track user behavior across channels; device fingerprinting provides a unique device identifier.

Device fingerprint ID: Consists of OS type, plugins, language, timezone, hardware IDs (IMEI, MAC, etc.) hashed into a fingerprint. Core challenges are ID validity (conflicts, drift) and service scalability/reliability.

Android device IDs: IMEI, IMSI, MAC, ANDROID_ID (unstable, can be changed).

iOS device IDs: IDFA, IDFV, UDID (deprecated), UUID, MAC (restricted).

Next‑generation fingerprinting: Uses multiple signals to generate a stable ID. Example: Alibaba's UMID, which aggregates IDs such as IMEI, IMSI, MAC, IDFA, YunosUUID, Google Advertising ID.

ID selection criteria: Minimize conflicts and drift, remain stable under permission limits, device re‑installations, and app reinstallations.

Architecture design: MySQL stores the main device table keyed by UMID (sharding required); HiStore provides column‑store OLAP for multi‑dimensional queries; Tair (or Redis) serves as a distributed key/value cache/persistent store for real‑time lookups.

Practical scenario: Advertising bidding strategies (OCPC, OCPA, OCPM) use device fingerprints to match clicks to conversions, enabling dynamic bid adjustment based on predicted conversion rates.

Data flow: Advertisers send click events with device IDs; Xianyu stores callbacks in Tair; upon user activation the device info is cleaned and merged, callbacks are triggered, and advertisers receive conversion data to optimize bids.

Conclusion: Device fingerprint services not only improve ad cost efficiency but also support deep linking, intelligent delivery, and lifecycle management, with future work focusing on further technical innovations for user growth.