Multi-Business Ranking Modeling in Meituan Search
Meituan Search tackles the multi‑business ranking challenge by introducing a quota‑allocation model (MQM) and a series of precise ranking models (MBN) that progressively incorporate sub‑networks, multi‑task learning and transformer‑based behavior sequences, delivering consistent CTR and purchase‑rate gains across food, hotel, travel and other services while outlining future work on feature utilization, sample‑imbalance mitigation and multi‑objective optimization.
Meituan Search serves a wide range of services (food delivery, in‑store dining, hotels, travel, entertainment, etc.). When a user searches a vague term such as “Wangjing”, the result page contains items from multiple business lines, creating a classic multi‑business mixed ranking problem.
The article outlines the challenges of multi‑business ranking: differing feature sensitivities across businesses, data imbalance between high‑frequency and low‑frequency services, and distinct business‑specific objectives.
Ranking pipeline : data → recall → ranking → display. The ranking layer consists of coarse ranking, multi‑path fusion, precise ranking, re‑ranking, and heterogeneous ranking.
Multi‑business quota model (MQM) – a quota‑allocation model that balances the proportion of candidates from each business. Two versions are described:
MQM‑V1 uses a single‑objective (click + order) probability per business and yields a +0.53% lift in overall CTR.
MQM‑V2 adds a second objective (cross‑business interaction) and a Transformer‑based behavior‑sequence module, achieving further improvements (e.g., +2% travel purchase rate).
Multi‑business precise ranking model (MBN) evolves through four versions:
MBN‑V1 introduces independent sub‑networks for low‑traffic businesses (e.g., hotels, travel) and combines their outputs with a weighted sum.
MBN‑V2 adds a food sub‑network and decouples quota output from the ranking model by learning weight generation from query/context features.
MBN‑V3 replaces handcrafted sub‑network inputs with an MMoE multi‑task learning layer and adds a business‑specific cross‑entropy loss, improving several business purchase rates.
MBN‑V4 adopts a CGC (single‑layer PLE) structure, yielding more stable expert weights and further gains in food business metrics.
Online experiments show consistent CTR and purchase‑rate improvements across versions, demonstrating the effectiveness of multi‑business modeling.
Future work includes better utilization of business‑specific features, addressing sample‑imbalance via transfer/meta‑learning, and multi‑objective optimization to align user experience with diverse business goals.
Signed-in readers can open the original source through BestHub's protected redirect.
This article has been distilled and summarized from source material, then republished for learning and reference. If you believe it infringes your rights, please contactand we will review it promptly.
Meituan Technology Team
Over 10,000 engineers powering China’s leading lifestyle services e‑commerce platform. Supporting hundreds of millions of consumers, millions of merchants across 2,000+ industries. This is the public channel for the tech teams behind Meituan, Dianping, Meituan Waimai, Meituan Select, and related services.
How this landed with the community
Was this worth your time?
0 Comments
Thoughtful readers leave field notes, pushback, and hard-won operational detail here.