Construction and Practical Application of a User Profile Tagging System

The presentation introduces the concept of a profile tag system, explaining why QuNar needs to consolidate disparate tag taxonomies from multiple business lines into a unified framework that supports strategic decision‑making.

It describes the sources of tag requirements, distinguishing between marketing‑risk control, internal business analysis, and user description needs, and outlines how these requirements drive the design of the tag hierarchy.

Tag classification is divided into business‑oriented categories and technical categories, with further subdivisions into statistical, rule‑based, and model‑based tags. The construction methods, update frequencies (hourly, daily, real‑time), and access patterns (offline storage in Redis/HBase versus online real‑time serving) are discussed.

The core platform, referred to as a CDP (Customer Data Platform), handles tag production, data analysis, business application, and effect evaluation, integrating with existing strategy platforms to enable end‑to‑end data‑driven marketing.

Algorithmic tag types are explored, including classification, recommendation, knowledge‑graph, causal inference, image processing, NLP, and look‑alike algorithms, with examples of how each is applied to user and item profiling.

Practical applications are presented: (1) marketing audience selection and expansion using tag‑driven segmentation, (2) business metric attribution analysis to diagnose performance issues, and (3) A/B experiment effectiveness analysis leveraging funnel models, tag identification, and decision‑tree diagnostics.

A Q&A section addresses common concerns such as the difference between user behavior and business logs, implementation of streaming tags with Flink/Spark, real‑time tag definitions, ID mapping strategies, and the role of large models in tag generation and knowledge‑graph construction.