NLP‑Driven Scenario Tagging and Experience Management Platform for Douyin App

Abstract Douyin receives massive daily user feedback; by applying an NLP‑based intelligent tagging model, the platform creates scenario‑specific tags, aggregates experience metrics from a business perspective, and uses word‑clustering to surface daily hotspots, enabling rapid issue identification and product experience enhancement.

Platform Background

With hundreds of millions of daily active users, Douyin faces two main challenges: feedback lacks scenario granularity, and extracting effective feedback for problem localization is difficult. The experience management platform was created to address these pain points by turning feedback into data‑driven insights for retention, growth, and reputation improvement.

Feedback Lifecycle

User feedback is reported from the client to the backend, stored in a database, and displayed on a console for various roles to process. The platform consists of two parts: a feedback workbench for tagging, filtering, and classification, and an experience management module that maps tags to scenarios, derives valuable indicators, and supports deep analysis and issue resolution.

NLP‑Empowered Scenario Tagging

Intelligent Feedback Tagging Model

The workbench provides viewing, replying, classification, and labeling of feedback, while maintaining customizable tag trees. To handle massive feedback, a pre‑training pipeline is used: domain‑adaptive pre‑training on feedback data, task‑adaptive pre‑training on business data, and fine‑tuning a classification model, which significantly improves tagging accuracy.

To mitigate long‑tail label imbalance, the model incorporates label‑transfer learning, multi‑label classification, and hierarchical classification to learn tree‑structured relationships among tags.

Business Scenario Tag Mapping

A visual console allows flexible configuration of business‑to‑tag mappings, supporting plug‑in style extensions. Tags follow a three‑level hierarchy (e.g., "Advertising" → "Ad complaints" → "Sell‑goods ad complaint"). This mapping links business scenarios such as "Douyin Lite" to relevant tags, enabling precise attribution of feedback.

Data‑Driven Problem Discovery

Scenario tags enable aggregation of feedback by business dimension, facilitating metric definition. The platform defines an "assistance rate" (feedback count per million DAU) as a primary experience indicator, expecting it to decline as issues are resolved. Additional metrics include top‑scenario feedback volume, change rate, and change amount.

Real‑Time Hot Word Clustering

A word‑cloud tool visualizes key phrases using intelligent segmentation, new‑word discovery, keyword extraction, and clustering algorithms, overcoming the limitations of traditional fine‑grained tokenization and providing richer insight into user concerns.

User Portrait Retrieval for Issue Analysis

By indexing raw feedback ("original voice"), the platform offers multi‑dimensional filters (gender, city, age, device, etc.) to build user portraits, helping pinpoint problematic segments such as Android‑specific issues and guiding targeted experience improvement plans.

Closed‑Loop Experience Management

Identified issues are fed back to product and R&D teams via a web‑based workflow that links feedback keywords, tags, or IDs to concrete improvement tasks, tracks progress, enforces permission controls, and records execution history, aiming to lower the assistance rate.

Platform Data Index Acceleration

To reduce query latency on massive data, the system employs offline pre‑processing for heavy‑weight metrics, periodic cache refreshes for hot queries, fallback data strategies for extreme cases, and a request flow that prioritizes cache, then pre‑processed data, and finally raw computation.

Conclusion

The experience management platform transforms user feedback into actionable, data‑driven insights through NLP‑enabled tagging, scenario mapping, metric quantification, and portrait analysis, supporting dozens of Douyin business scenarios and driving continuous product experience improvement.