DynamicPO: Dynamic Preference Optimization for Recommendation

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Hu, Xingyu, Zhang, Kai, Wu, Jiancan, Wang, Shuli, Wang, Chi, Chen, Wenshuai, Zhu, Yinhua, Wang, Haitao, Wang, Xingxing, Wang, Xiang
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866914524131491840
author Hu, Xingyu
Zhang, Kai
Wu, Jiancan
Wang, Shuli
Wang, Chi
Chen, Wenshuai
Zhu, Yinhua
Wang, Haitao
Wang, Xingxing
Wang, Xiang
author_facet Hu, Xingyu
Zhang, Kai
Wu, Jiancan
Wang, Shuli
Wang, Chi
Chen, Wenshuai
Zhu, Yinhua
Wang, Haitao
Wang, Xingxing
Wang, Xiang
contents In large language model (LLM)-based recommendation systems, direct preference optimization (DPO) effectively aligns recommendations with user preferences, requiring multi-negative objective functions to leverage abundant implicit-feedback negatives and sharpen preference boundaries. However, our empirical analyses reveal a counterintuitive phenomenon, preference optimization collapse, where increasing the number of negative samples can lead to performance degradation despite a continuously decreasing training loss. We further theoretically demonstrate that this collapse arises from gradient suppression, caused by the dominance of easily discriminable negatives over boundary-critical negatives that truly define user preference boundaries. As a result, boundary-relevant signals are under-optimized, weakening the model's decision boundary. Motivated by these observations, we propose DynamicPO (Dynamic Preference Optimization), a lightweight and plug-and-play framework comprising two adaptive mechanisms: Dynamic Boundary Negative Selection, which identifies and prioritizes informative negatives near the model's decision boundary, and Dual-Margin Dynamic beta Adjustment, which calibrates optimization strength per sample according to boundary ambiguity. Extensive experiments on three public datasets show that DynamicPO effectively prevents optimization collapse and improves recommendation accuracy on multi-negative preference optimization methods, with negligible computational overhead. Our code and datasets are available at https://github.com/xingyuHuxingyu/DynamicPO.
format Preprint
id arxiv_https___arxiv_org_abs_2605_00327
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle DynamicPO: Dynamic Preference Optimization for Recommendation
Hu, Xingyu
Zhang, Kai
Wu, Jiancan
Wang, Shuli
Wang, Chi
Chen, Wenshuai
Zhu, Yinhua
Wang, Haitao
Wang, Xingxing
Wang, Xiang
Information Retrieval
Artificial Intelligence
In large language model (LLM)-based recommendation systems, direct preference optimization (DPO) effectively aligns recommendations with user preferences, requiring multi-negative objective functions to leverage abundant implicit-feedback negatives and sharpen preference boundaries. However, our empirical analyses reveal a counterintuitive phenomenon, preference optimization collapse, where increasing the number of negative samples can lead to performance degradation despite a continuously decreasing training loss. We further theoretically demonstrate that this collapse arises from gradient suppression, caused by the dominance of easily discriminable negatives over boundary-critical negatives that truly define user preference boundaries. As a result, boundary-relevant signals are under-optimized, weakening the model's decision boundary. Motivated by these observations, we propose DynamicPO (Dynamic Preference Optimization), a lightweight and plug-and-play framework comprising two adaptive mechanisms: Dynamic Boundary Negative Selection, which identifies and prioritizes informative negatives near the model's decision boundary, and Dual-Margin Dynamic beta Adjustment, which calibrates optimization strength per sample according to boundary ambiguity. Extensive experiments on three public datasets show that DynamicPO effectively prevents optimization collapse and improves recommendation accuracy on multi-negative preference optimization methods, with negligible computational overhead. Our code and datasets are available at https://github.com/xingyuHuxingyu/DynamicPO.
title DynamicPO: Dynamic Preference Optimization for Recommendation
topic Information Retrieval
Artificial Intelligence
url https://arxiv.org/abs/2605.00327