You Only Evaluate Once: A Tree-based Rerank Method at Meituan

2025-08-20

AI Summary

Existing re-ranking methods for recommender systems suffer from inconsistency: generic search units (GSUs) frequently overlook high-value candidate lists generated by exact search units (ESUs). To address this, we propose YOLOR—a tree-structured, single-stage, list-wise re-ranker that eliminates the coarse-filtering stage of conventional two-stage paradigms and directly models full-candidate permutations. Its core innovations are the Tree-structured Context Extraction Module (TCEM), which hierarchically aggregates multi-scale contextual features, and the Context Caching Module (CCM), enabling efficient feature reuse across candidate permutations. Evaluated on multiple public and industrial datasets, YOLOR achieves significant improvements in both re-ranking accuracy and inference efficiency. It has been successfully deployed in Meituan’s food delivery platform, demonstrating strong practical effectiveness and scalability.

Technology Category

• Machine Learning: Learning Preferences or Rankings
• Search and Optimization: Learning to Search
• Data Mining & Knowledge Management: Recommender Systems

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Graph Algorithms and Modeling for the Web: Querying, indexing, and retrieval in Web-related graphs

Abstract

Reranking plays a crucial role in modern recommender systems by capturing the mutual influences within the list. Due to the inherent challenges of combinatorial search spaces, most methods adopt a two-stage search paradigm: a simple General Search Unit (GSU) efficiently reduces the candidate space, and an Exact Search Unit (ESU) effectively selects the optimal sequence. These methods essentially involve making trade-offs between effectiveness and efficiency, while suffering from a severe extbf{inconsistency problem}, that is, the GSU often misses high-value lists from ESU. To address this problem, we propose YOLOR, a one-stage reranking method that removes the GSU while retaining only the ESU. Specifically, YOLOR includes: (1) a Tree-based Context Extraction Module (TCEM) that hierarchically aggregates multi-scale contextual features to achieve "list-level effectiveness", and (2) a Context Cache Module (CCM) that enables efficient feature reuse across candidate permutations to achieve "permutation-level efficiency". Extensive experiments across public and industry datasets validate YOLOR's performance, and we have successfully deployed YOLOR on the Meituan food delivery platform.

Problem

• Addresses inconsistency between candidate generation and reranking stages
• Eliminates separate candidate reduction step in recommender systems
• Improves both effectiveness and efficiency in list reranking

Innovation

• One-stage reranking method removes GSU
• Tree-based module aggregates contextual features hierarchically
• Cache module enables efficient feature reuse