Zhihu Bridge Platform: Internal Marketing Architecture, Challenges, and Optimizations

Introduction

The Bridge Platform is a one‑stop solution for content, user, and creator management, offering capabilities such as filtering, packaging, analysis, monitoring, marketing, delivery, and intervention, with a focus on internal marketing.

Background and Origin

Internal marketing aims to increase domain DAU, content production, creator growth, and revenue through targeted activities and incentives.

Business Logic Overview

Three core elements—reach channel, target audience, and incentive/marketing method—form the internal marketing loop that moves users between lifecycle stages.

Product Function Concepts

1. Customer Data Platform (CDP)

CDP aggregates multi‑source user data to support audience segmentation and insight generation.

Audience Segmentation : Define user groups by geography, behavior, interests, etc.

Audience Insight : Analyze segmented groups for behavior patterns and preferences.

2. Activity Platform

Provides activity design, execution, and analysis, including activity building, gameplay design, prize distribution, and landing page creation.

3. Internal Delivery Platform

Handles resource‑slot delivery, feed delivery, and proactive push notifications across APP and PC touchpoints.

Business Architecture

Macro Architecture

Four layers: Business Capability, Shared Service, Business Model, and External Access.

Subsystems

Includes Activity Platform, CDP, Internal Delivery Platform, User Behavior System, Content Event Center, and Fact Sharing Service.

Challenges and Breakthroughs

1. Activity Platform

Implemented reliable event queues with idempotency and TCC distributed transactions to ensure consistency across task completion, point awarding, and prize delivery.

2. Internal Delivery Platform

Unified delivery center reduces coordination overhead; introduced automated audience targeting via strategy rules and recommendation algorithms.

3. CDP Query Performance

Optimized by converting AND/OR/NOT to bitmap functions, applying bitmap indexing, and partitioning data to reduce network I/O; later introduced a divide‑and‑conquer approach and function‑level operator enhancements.

4. Shared Fact Service Import

Reduced offline batch load time from 8+ hours to ~55 minutes (9× speedup) by leveraging Spark, broker load, and CPU‑intensive optimizations.

5. Vectorized Query in Doris

Enabled vectorized execution, binary bitmap return, and increased thread pool size, achieving 2‑14× query speed improvements across various workloads.

Future Outlook

Roadmap moves from marketing digitization to automation (MA) and finally to intelligent marketing, integrating AB testing, automated workflow orchestration, and AI‑driven decision making.