Building and Operating a User Portrait Platform: Architecture, Practices, and Case Studies from Kuakan Comics

The portrait platform is a core middle‑platform product widely used across business lines for personalized recommendation, fine‑grained operations, precise marketing, and acquisition promotion.

Product Architecture : The platform serves the business layer with functions such as personalized recommendation, fine‑grained operation, precise marketing, and acquisition promotion. Core service capabilities include crowd selection, crowd computation, crowd insight, crowd‑package management, tag management, and API services.

Platform Positioning : Two deployment models are discussed – (1) business‑side self‑built platforms for large, data‑mature lines, and (2) a middle‑platform portrait platform built by the data‑center to serve multiple business lines, emphasizing decoupling, data standardization, and extensibility.

Construction Experience :

Device‑ID (KKDID) strategy: collect all available IDs (IMEI, IDFA, OAID, etc.), maintain mapping and change history, and generate an internal unique identifier to improve coverage and accuracy.

Data‑warehouse modeling: three‑layer approach – data‑domain layer (DWD/DWS) for raw business data, tag‑calculation and mining layer, and portrait‑theme layer (DM/APP) for tag storage and management.

Multi‑business tag fusion: define naming, accuracy, coverage, and value standards; automatically merge custom tags from various lines into the portrait theme.

Composite tags: support complex, multi‑dimensional tag definitions (e.g., 90‑day consumption with conditional filters) to enable flexible analysis and reduce development cost.

Application Cases :

Precision‑marketing loop: select crowd packages via the portrait backend, push targeted campaigns through marketing APIs, collect BI reports, and iterate based on performance.

Content precise distribution: match comic genres with user preference crowds, configure delivery slots, collect results, and refine strategies using BI feedback.

Summary and Outlook : Emphasizes the importance of a middle‑platform portrait solution, the need for robust big‑data and data‑warehouse capabilities, and outlines future work – systematic tag management, enhanced service APIs, and tighter integration of real‑time and offline portraits.

Q&A Highlights :

Device‑ID generation rules and privacy considerations.

Balancing user disturbance with fine‑grained operation.

Key points for building a portrait platform from 0 to 1, including data‑modeling, tag‑fusion, and MVP selection.

Limitations of Cookie‑ID for H5 pages and the preference for device‑level IDs.