Probabilistic Residual Learning for Online Recommendations

2026-07-22

AI Summary

This work addresses the challenges posed by the black-box nature and high computational complexity of existing deep recommender systems, which hinder systematic performance optimization. To overcome these limitations, the authors propose a Probabilistic Residual Learning (PRL) framework that models the prediction residuals of a base recommender through a causal Bayesian model. By integrating user clustering, domain confounder identification, and do-calculus, PRL establishes a plug-and-play residual refinement mechanism that requires no modification to the original model architecture. Extensive experiments demonstrate that PRL consistently enhances recommendation performance across multiple state-of-the-art deep recommender systems. Moreover, the framework automatically uncovers semantically meaningful user segments, thereby offering both performance gains and improved interpretability.

Abstract

Modern recommender systems are typically based on deep learning (DL) models, where a dense encoder learns representations of users and items. As a result, these systems often suffer from the black-box nature and computational complexity of the underlying models, making it difficult to systematically enhance their recommendation capabilities. To address this problem, we propose Probabilistic Residual Learning (PRL), a causal Bayesian recommendation model that models the residual between ground-truth and base predictions, enabling targeted refinement of existing systems. Specifically, PRL (1) probabilistically groups users for localized residual modeling, (2) models domain-level confounders that influence user and item representations, and (3) aggregates cluster-specific residual predictions over the confounders using do-calculus. Experiments demonstrate that our plug-and-play PRL is compatible with various base deep learning recommender systems, improving their performance while automatically discovering meaningful user clusters.

Problem

• recommender systems
• black-box models
• computational complexity
• recommendation enhancement

Innovation

• Probabilistic Residual Learning
• Causal Recommendation
• Do-calculus
• User Clustering
• Plug-and-play Enhancement