Learning Cascade Ranking as One Network

2025-03-12

AI Summary

In large cascaded ranking systems, conventional single-stage training overlooks inter-stage interactions and misaligns with the end-to-end recall objective. To address this, we propose LCRON—a framework for end-to-end differentiable cascaded ranking. Its key contributions are: (1) the first derivation of a differentiable surrogate loss grounded in a lower bound of the ground-truth probability of being selected by the cascade, enabling joint optimization aligned with the end-to-end recall goal; and (2) stage-aware auxiliary losses explicitly modeling inter-stage collaboration. LCRON constructs a fully differentiable cascaded model that supports gradient backpropagation and cooperative training across stages. Extensive experiments on public benchmarks and industrial deployment scenarios demonstrate that LCRON significantly outperforms state-of-the-art methods—including RankFlow and FS-LTR—achieving superior top-k selection accuracy and enhanced system robustness.

Technology Category

• Machine Learning: Learning Preferences or Rankings
• Search and Optimization: Learning to Search
• Data Mining & Knowledge Management: Conversational Systems for Recommendation & Retrieval

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
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Cascade Ranking is a prevalent architecture in large-scale top-k selection systems like recommendation and advertising platforms. Traditional training methods focus on single-stage optimization, neglecting interactions between stages. Recent advances such as RankFlow and FS-LTR have introduced interaction-aware training paradigms but still struggle to 1) align training objectives with the goal of the entire cascade ranking (i.e., end-to-end recall) and 2) learn effective collaboration patterns for different stages. To address these challenges, we propose LCRON, which introduces a novel surrogate loss function derived from the lower bound probability that ground truth items are selected by cascade ranking, ensuring alignment with the overall objective of the system. According to the properties of the derived bound, we further design an auxiliary loss for each stage to drive the reduction of this bound, leading to a more robust and effective top-k selection. LCRON enables end-to-end training of the entire cascade ranking system as a unified network. Experimental results demonstrate that LCRON achieves significant improvement over existing methods on public benchmarks and industrial applications, addressing key limitations in cascade ranking training and significantly enhancing system performance.

Problem

• Aligns training objectives with cascade ranking's end-to-end recall.
• Learns effective collaboration patterns across different ranking stages.
• Improves top-k selection robustness and effectiveness in cascade systems.

Innovation

• Introduces LCRON for cascade ranking optimization
• Uses surrogate loss for end-to-end recall alignment
• Designs auxiliary loss for robust top-k selection