Causal Negative Sampling via Diffusion Model for Out-of-Distribution Recommendation

2025-08-10

AI Summary

In recommender systems, heuristic negative sampling is susceptible to environmental confounders—such as exposure bias and item popularity—leading to spurious hard negative samples (FHNS) that induce false correlations and impair out-of-distribution generalization. To address this, we propose Causal Diffusion Negative Sampling (CDNS): a novel framework that leverages a conditional diffusion model to generate deconfounded negative samples in the latent space, augmented with a causal regularization term that explicitly mitigates confounder influence. CDNS is the first approach to unify causal inference and diffusion modeling for negative sampling, jointly achieving generative debiasing and causal disentanglement. Extensive experiments across four distribution shift scenarios demonstrate that CDNS consistently outperforms state-of-the-art methods by an average of 13.96%, significantly enhancing model robustness and generalization capability.

Technology Category

• Machine Learning: Causal Learning
• Computer Vision: Diffusion Models for Vision
• Reasoning under Uncertainty: Causality

Application Category

• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs

Abstract

Heuristic negative sampling enhances recommendation performance by selecting negative samples of varying hardness levels from predefined candidate pools to guide the model toward learning more accurate decision boundaries. However, our empirical and theoretical analyses reveal that unobserved environmental confounders (e.g., exposure or popularity biases) in candidate pools may cause heuristic sampling methods to introduce false hard negatives (FHNS). These misleading samples can encourage the model to learn spurious correlations induced by such confounders, ultimately compromising its generalization ability under distribution shifts. To address this issue, we propose a novel method named Causal Negative Sampling via Diffusion (CNSDiff). By synthesizing negative samples in the latent space via a conditional diffusion process, CNSDiff avoids the bias introduced by predefined candidate pools and thus reduces the likelihood of generating FHNS. Moreover, it incorporates a causal regularization term to explicitly mitigate the influence of environmental confounders during the negative sampling process, leading to robust negatives that promote out-of-distribution (OOD) generalization. Comprehensive experiments under four representative distribution shift scenarios demonstrate that CNSDiff achieves an average improvement of 13.96% across all evaluation metrics compared to state-of-the-art baselines, verifying its effectiveness and robustness in OOD recommendation tasks.

Problem

• Addresses false hard negatives from environmental confounders in recommendation
• Reduces spurious correlations caused by exposure and popularity biases
• Improves out-of-distribution generalization in recommendation systems

Innovation

• Synthesizes negative samples via diffusion process
• Incorporates causal regularization to mitigate confounders
• Generates robust negatives for OOD generalization