Addressing Missing Data Issue for Diffusion-based Recommendation

TL;DR

This work tackles missing data in diffusion-based sequential recommendations by introducing Diffusion with Dual-side Thompson Sampling (TDM). TDM simulates extra missing guidance signals using local continuity and global sequence stability to edit observed sequences, enabling diffusion models to extrapolate and learn consistency under missing-data perturbations. The approach combines Dual-side Thompson Sampling with Denoising Diffusion Implicit Models (DDIM) to accelerate generation and achieve robustness, supported by theoretical extrapolation/consistency arguments and extensive experiments across multiple datasets. Results show substantial gains over state-of-the-art baselines and demonstrate the method's robustness to varying missing-data levels and sequence lengths, with beneficial generalization to non-diffusion recommendations.

Abstract

Diffusion models have shown significant potential in generating oracle items that best match user preference with guidance from user historical interaction sequences. However, the quality of guidance is often compromised by unpredictable missing data in observed sequence, leading to suboptimal item generation. Since missing data is uncertain in both occurrence and content, recovering it is impractical and may introduce additional errors. To tackle this challenge, we propose a novel dual-side Thompson sampling-based Diffusion Model (TDM), which simulates extra missing data in the guidance signals and allows diffusion models to handle existing missing data through extrapolation. To preserve user preference evolution in sequences despite extra missing data, we introduce Dual-side Thompson Sampling to implement simulation with two probability models, sampling by exploiting user preference from both item continuity and sequence stability. TDM strategically removes items from sequences based on dual-side Thompson sampling and treats these edited sequences as guidance for diffusion models, enhancing models' robustness to missing data through consistency regularization. Additionally, to enhance the generation efficiency, TDM is implemented under the denoising diffusion implicit models to accelerate the reverse process. Extensive experiments and theoretical analysis validate the effectiveness of TDM in addressing missing data in sequential recommendations.

Figures (6)

• Figure 1: Phenomenon of uncertain missing data in sequences and the method comparison to address it. The green curve represents the evolution of user preference over time.
• Figure 2: The overview of the TDM framework, which simulates extra missing data with DTS in the guidance signals, achieving diffusion models' consistency regularization and extrapolating to address the existing missing data.
• Figure 3: Sensitivity of TDM to the hyperparameter of $\lambda_1$ on multiple datasets, demonstrating the proportion of edited sequences. The "random" represents the variant "w/o GL" of TDM.
• Figure 4: Sensitivity of TDM to the hyperparameter of $\lambda_2$ on multiple datasets, demonstrating the proportion of removed items. The "random" represents the variant "w/o GL" of TDM.
• Figure 5: Performance of TDM on synthetic datasets with different missing ratios.