On Efficiency-Effectiveness Trade-off of Diffusion-based Recommenders

TL;DR

This paper introduces TA-Rec, a two-stage framework that improves the efficiency and effectiveness of diffusion-based recommenders by smoothing the denoising process and aligning it with user preferences, reducing discretization errors.

Contribution

The paper proposes TA-Rec, a novel two-stage approach with Temporal Consistency Regularization and Adaptive Preference Alignment to balance efficiency and effectiveness in diffusion recommendation models.

Findings

TA-Rec reduces discretization errors in diffusion recommenders.

TA-Rec achieves faster one-step generation without loss of accuracy.

Experiments show improved recommendation performance and efficiency.

Abstract

Diffusion models have emerged as a powerful paradigm for generative sequential recommendation, which typically generate next items to recommend guided by user interaction histories with a multi-step denoising process. However, the multi-step process relies on discrete approximations, introducing discretization error that creates a trade-off between computational efficiency and recommendation effectiveness. To address this trade-off, we propose TA-Rec, a two-stage framework that achieves one-step generation by smoothing the denoising function during pretraining while alleviating trajectory deviation by aligning with user preferences during fine-tuning. Specifically, to improve the efficiency without sacrificing the recommendation performance, TA-Rec pretrains the denoising model with Temporal Consistency Regularization (TCR), enforcing the consistency between the denoising results across adjacent steps. Thus, we can smooth the denoising function to map the noise as oracle items in one step with bounded error. To further enhance effectiveness, TA-Rec introduces Adaptive Preference Alignment (APA) that aligns the denoising process with user preference adaptively based on preference pair similarity and timesteps. Extensive experiments prove that TA-Rec's two-stage objective effectively mitigates the discretization errors-induced trade-off, enhancing both efficiency and effectiveness of diffusion-based recommenders.