RAD-DPO: Robust Adaptive Denoising Direct Preference Optimization For Generative Retrieval In E-commerce

Abstract

arXiv:2602.23964v2 Announce Type: replace Abstract: Generative Retrieval (GR) is rapidly transforming e-commerce search by replacing traditional multi-stage pipelines with the autoregressive decoding of structured Semantic IDs (SIDs). Despite this architectural efficiency, aligning GR models with nuanced, real-world user preferences remains a critical challenge. While Direct Preference Optimization (DPO) offers an efficient alignment solution, its direct application to structured SIDs suffers from three limitations: (i) it penalizes shared hierarchical prefixes, causing gradient conflicts; (ii) it is vulnerable to noisy pseudo-negatives from implicit feedback; and (iii) in multi-label queries with multiple relevant items, it exacerbates a probability "squeezing effect" among valid candidates. To address these issues, we propose RAD-DPO, which introduces token-level gradient detachment to protect prefix structures, similarity-based dynamic reward weighting to mitigate label noise, and

Stats

• citations0
• S2 citations—
• github stars0
• HF likes0
• heat score0.00
• arxiv keychen2026rad

Related papers

• Differentiable Geometric Indexing For End-to-end Generative Retrieval (2026)0.00
• Lightweight And Direct Document Relevance Optimization For Generative Information Retrieval (2025)4.52
• RADAR: Recall Augmentation Through Deferred Asynchronous Retrieval (2025)2.26
• Generative Retrieval Meets Multi-graded Relevance (2024)2.26
• Planning Ahead In Generative Retrieval: Guiding Autoregressive Generation Through Simultaneous Decoding (2024)8.82
• Retrieval-grpo: A Multi-objective Reinforcement Learning Framework For Dense Retrieval In Taobao Search (2025)0.00
• Breaking The Hourglass Phenomenon Of Residual Quantization: Enhancing The Upper Bound Of Generative Retrieval (2024)4.52
• Does Generative Retrieval Overcome The Limitations Of Dense Retrieval? (2025)0.00