Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Enhancing guidance for missing data in diffusion-based sequential recommendation

Qilong Yan, Yifei Xing, Dugang Liu, Jingpu Duan, Jian Yin

TL;DR

This paper tackles missing data in diffusion-guided sequential recommendation by introducing CARD, a model that amplifies signals from critical turning-point items while suppressing noise. It combines a stability-based routing strategy with a counterfactual attention mechanism to create a high-quality, dynamically re-weighted guidance signal for the diffusion process. The approach yields state-of-the-art recommendation accuracy on real-world datasets and improves training efficiency through selective application of intensive computations. Overall, CARD provides a robust, scalable solution for principled guidance in generative recommender systems and is accompanied by publicly available code.

Abstract

Contemporary sequential recommendation methods are becoming more complex, shifting from classification to a diffusion-guided generative paradigm. However, the quality of guidance in the form of user information is often compromised by missing data in the observed sequences, leading to suboptimal generation quality. Existing methods address this by removing locally similar items, but overlook ``critical turning points'' in user interest, which are crucial for accurately predicting subsequent user intent. To address this, we propose a novel Counterfactual Attention Regulation Diffusion model (CARD), which focuses on amplifying the signal from key interest-turning-point items while concurrently identifying and suppressing noise within the user sequence. CARD consists of (1) a Dual-side Thompson Sampling method to identify sequences undergoing significant interest shift, and (2) a counterfactual attention mechanism for these sequences to quantify the importance of each item. In this manner, CARD provides the diffusion model with a high-quality guidance signal composed of dynamically re-weighted interaction vectors to enable effective generation. Experiments show our method works well on real-world data without being computationally expensive. Our code is available at https://github.com/yanqilong3321/CARD.

Enhancing guidance for missing data in diffusion-based sequential recommendation

TL;DR

This paper tackles missing data in diffusion-guided sequential recommendation by introducing CARD, a model that amplifies signals from critical turning-point items while suppressing noise. It combines a stability-based routing strategy with a counterfactual attention mechanism to create a high-quality, dynamically re-weighted guidance signal for the diffusion process. The approach yields state-of-the-art recommendation accuracy on real-world datasets and improves training efficiency through selective application of intensive computations. Overall, CARD provides a robust, scalable solution for principled guidance in generative recommender systems and is accompanied by publicly available code.

Abstract

Contemporary sequential recommendation methods are becoming more complex, shifting from classification to a diffusion-guided generative paradigm. However, the quality of guidance in the form of user information is often compromised by missing data in the observed sequences, leading to suboptimal generation quality. Existing methods address this by removing locally similar items, but overlook ``critical turning points'' in user interest, which are crucial for accurately predicting subsequent user intent. To address this, we propose a novel Counterfactual Attention Regulation Diffusion model (CARD), which focuses on amplifying the signal from key interest-turning-point items while concurrently identifying and suppressing noise within the user sequence. CARD consists of (1) a Dual-side Thompson Sampling method to identify sequences undergoing significant interest shift, and (2) a counterfactual attention mechanism for these sequences to quantify the importance of each item. In this manner, CARD provides the diffusion model with a high-quality guidance signal composed of dynamically re-weighted interaction vectors to enable effective generation. Experiments show our method works well on real-world data without being computationally expensive. Our code is available at https://github.com/yanqilong3321/CARD.
Paper Structure (12 sections, 6 equations, 2 figures, 3 tables)

This paper contains 12 sections, 6 equations, 2 figures, 3 tables.

Table of Contents

  1. Introduction
  2. METHODOLOGY
  3. Preliminaries: Diffusion Models for Recommendation
  4. Dynamic Guidance Optimization via Routing
  5. Stability Evaluation.
  6. Processing of Low-Stability Sequences.
  7. Guidance Vector Generation.
  8. Model Training and Inference
  9. EXPERIMENTS
  10. Experimental Setup
  11. Results
  12. Conclusion

Figures (2)

  • Figure 1: Comparisons of existing methods for missing data: (a) Recovering-based methods risk introducing erroneous items. (b) Simulating missingness by removing items based on local continuity. (c) Our method by re-weighting items based on their predictive importance
  • Figure 2: The overall architecture follows a three-step process: (1) it first evaluates sequence stability to determine a processing path; (2) high-stability sequences are then simplified via DTS, while (3) low-stability ones are re-weighted using our counterfactual attention. The optimized sequence is then encoded to provide guidance for the diffusion model.