DynamicPO: Dynamic Preference Optimization for Recommendation

2026-04-30

AI Summary

This work addresses a critical yet previously overlooked issue in multi-negative-sample preference optimization: easily distinguishable negative samples dominate the training process, leading to gradient suppression and weakened boundary signals, which ultimately cause optimization collapse. The study is the first to uncover the gradient-level mechanisms underlying this phenomenon and proposes a lightweight, plug-and-play framework to mitigate it. By dynamically sampling hard negative examples near the decision boundary, incorporating a dual-margin design, and introducing a sample-level adaptive beta adjustment mechanism, the method sharpens boundary learning without significant computational overhead. Integrated within the DPO framework, the approach consistently achieves substantial gains in recommendation accuracy across three public benchmarks while effectively preventing optimization collapse.

Abstract

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.

Problem

• preference optimization collapse
• gradient suppression
• negative sampling
• decision boundary
• recommendation systems

Innovation

• Preference Optimization Collapse
• Gradient Suppression
• Dynamic Boundary Negative Selection
• Dual-Margin Dynamic Beta Adjustment
• Multi-negative DPO