Building User Profiles: From Zero to One and Scaling to Hundreds

User profiling is a structured way to capture extensive user information—both self‑reported and behaviorally collected—into a hierarchical tag tree that serves as a comprehensive user dossier for product, marketing, and algorithmic applications.

1. Building User Profiles from 0 to 1

A mature profile system contains hundreds of tags organized into eight dimensions: basic attributes, platform attributes, behavior attributes, product preferences, interest preferences, sensitivity, consumption attributes, and lifecycle/value. The tag tree must be highly inclusive (high generalization) and easily extensible (strong expandability) to accommodate future business needs.

Basic attributes are static, directly sourced from user‑provided data (e.g., age, gender, ID). Platform attributes are derived from behavioral data via algorithms (e.g., platform‑age, preference probabilities). The article contrasts data sources, calculation logic, and tag formats, illustrating with gender tags such as gender_female (basic) versus gender_female_0.80 (platform).

Behavior attributes record granular actions (login, click, purchase) and are organized by product, module, action, and time. Product preference tags capture affinity for specific products or channels, while interest preference tags describe brand, category, or label affinities with confidence scores. Sensitivity tags measure user responsiveness to discounts or promotions, and consumption tags quantify frequency, amount, and recent activity. Lifecycle tags label users as new, growing, mature, declining, or churned, and value tags (e.g., RFM) assess overall contribution.

2. Scaling User Profiles from 1 to 100

The second stage focuses on expanding the tag set to meet concrete business scenarios. It starts by clarifying the purpose of the profile—statistics, targeted marketing, or algorithmic features—and then iteratively designs tags that address each need.

For statistical insight, tags such as age, gender, and city are extracted directly from basic attributes. For precision marketing, tags like user_lifecycle_churn , discount_sensitivity , and push‑interaction metrics are added to enable accurate audience segmentation and efficient resource allocation.

In algorithmic contexts (e.g., recommendation recall), profile tags such as brand, category, and label preferences are used to retrieve candidate items from pre‑ranked lists, forming the initial pool for downstream ranking models.

The article concludes with a practical case study: using the profile to identify churned users, assess their discount sensitivity, and prioritize push notifications, illustrating how a well‑designed tag hierarchy translates into measurable business impact.