Design and Implementation of Baidu's Commodity Promotion System

This article introduces the design and implementation of Baidu's commodity promotion system, which was built starting in 2020 to support e-commerce scenarios in Baijiahao and live streaming. The system connects B-end merchants and C-end authors/streamers, providing convenient drop-shipping processes and simple shopping experiences for users.

The promotion system revolves around three core roles: traffic owners (authors/streamers), advertisers (merchants), and users (consumers). Traffic owners help increase product exposure and clicks, earning commissions when users purchase through their links. Advertisers provide products for promotion and pay commissions. The system uses CPS (Cost Per Sales) as the primary billing method, charging based on actual sales volume.

The system architecture includes three user interfaces: a mini-program B-end for advertisers to register products, a selection interface for traffic owners in Baijiahao and live streaming, and a user purchase interface. Five core services support these interfaces: mounting service, promotion service, transaction system, settlement system, and asset system.

Key technical implementations include:

1. Dynamic library-based promotion data reporting: Using Baidu's mini-program dynamic library framework to embed promotion tracking components that can silently update and provide consistent data collection across all Baidu ecosystem apps.

2. Database batch write optimization: Implementing asynchronous batch writes for promotion data to improve system throughput, as real-time queries are not required for this type of data.

The system addresses challenges such as providing unified technical solutions across multiple Baidu apps, effectively collecting user promotion behavior across diverse pages, and integrating with existing order and settlement systems. The design follows high cohesion and low coupling principles, with clear separation between business logic, access layers, and service layers.

The article concludes by emphasizing that building such a system requires cross-team collaboration and attention to both broad architectural considerations and detailed implementation aspects. The system leverages existing transaction platform capabilities to accelerate development while maintaining scalability and performance.