Architecture of a Community Advertising Placement System

The article outlines the design and implementation of an advertising placement system used in the Xianyu community, describing its four core elements—recruitment, selection, construction, and delivery—and how they map to the 4W1H model (What, When, How, Where, Who).

It defines a resource slot as a display coordinate on a page, detailing its key attributes such as ID, status, arbitration rules, material templates, page URL, fatigue control, and traffic limits.

Material templates standardize various content types (e.g., products, posts, banners) to simplify creation, while a material instance is generated by filling a chosen template with specific data.

Advertising tasks are the central unit, comprising an ID, schedule, associated resource slots, material, weight, target audience, platform constraints, region, fatigue settings, traffic caps, and custom tag rules.

The system architecture is divided into three layers: the front‑end business layer (operations console and consumer‑facing scenes), the business service layer (tenant management, template design, validation, search, preview, reporting, approval, rule definition, and targeting), and the service foundation layer (distributed cache, messaging, database access, search, stream processing, audience and region services).

Structured data (tasks, slots, templates, materials) are stored in MySQL, while task retrieval leverages Opensearch. Real‑time metrics are persisted in Lindorm, and high‑performance key‑value caching uses Tair.

Data‑driven feedback loops enable real‑time monitoring of task exposure and clicks, supporting traffic strategies such as slot‑level caps, task‑level caps, race‑horse mechanisms, and cold‑start support for new tasks.

Case studies include an “irregular guide” growth project that increased conversion by 56.1% and a “hot chat” module that boosted user retention.

The outlook highlights future work: integrating algorithmic matching, real‑time behavior intervention, and historical data accumulation to refine the system.