How Meilishuo Personalizes Fashion: Inside Its AI‑Driven Recommendation Engine

Gu Hao, senior architect at Meilishuo, leads personalization and advertising monetization technologies. With a background from Baidu's ad backend, he discusses how fragmented mobile user attention challenges the delivery of attractive content.

Meilishuo, China’s largest fast‑fashion discovery platform, has evolved from a social e‑commerce guide to a full‑stack vertical fashion marketplace, now hosting over 15,000 merchants and billions of RMB in GMV.

User Value and Product Scenarios

The platform aims to respect limited user attention by presenting timely, relevant fashion items across push notifications, recommendation pages, search results, and product detail pages.

Technical Challenges

Data collection, monitoring, and processing

Unified user representation, real‑time feedback, distributed processing

User understanding and representation

Personalized candidate generation and ranking

Interaction design and evaluation/optimization

Item Information and User Behavior

Example product attributes (e.g., dress category, color, style, material) are used together with short‑term preferences (category, shop, style) and long‑term preferences (spending power, color, size) to model user taste.

Personalization Framework

The system combines item clustering, scenario adaptation, collaborative filtering / matrix factorization, p‑value estimation, real‑time feedback, attribute expansion, and goal fusion to improve recall, precision, and user engagement.

Personalized Push and Recommendation

Push messages are selected by content, timing, landing page, and flexible frequency to increase app opens while reducing fatigue. Evaluation metrics include UV and GMV.

Practical Architecture

Before solving a problem, the team defines it: personalization is a system‑level effort that respects users, goes beyond KPI, and delivers a sense of being understood.