Constraint-Aware Generative Re-ranking for Multi-Objective Optimization in Advertising Feeds

TL;DR

This paper introduces a unified, constraint-aware generative reranking framework for advertising feeds that improves revenue and user engagement while satisfying strict latency and constraint requirements.

Contribution

It presents a novel unified neural model that combines sequence generation and reward estimation, with constraint-aware reward pruning for efficient constrained optimization.

Findings

Improves platform revenue and user engagement in large-scale industrial feeds.

Reduces inference latency compared to prior generative ranking methods.

Effectively handles constraints within the decoding process, enabling real-time deployment.

Abstract

Optimizing reranking in advertising feeds is a constrained combinatorial problem, requiring simultaneous maximization of platform revenue and preservation of user experience. Recent generative ranking methods enable listwise optimization via autoregressive decoding, but their deployment is hindered by high inference latency and limited constraint handling. We propose a constraint-aware generative reranking framework that transforms constrained optimization into bounded neural decoding. Unlike prior approaches that separate generator and evaluator models, our framework unifies sequence generation and reward estimation into a single network. We further introduce constraint-aware reward pruning, integrating constraint satisfaction directly into decoding to efficiently generate optimal sequences. Experiments on large-scale industrial feeds and online A/B tests show that our method improves revenue and user engagement while meeting strict latency requirements, providing an efficient neural solution for constrained listwise optimization.