Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Denoised Recommendation Model with Collaborative Signal Decoupling

Zefeng Li, Ning Yang

TL;DR

The paper tackles noise in user-item interaction graphs that degrades collaborative filtering performance. It introduces DRCSD, a GNN-based framework with a Collaborative Signal Decoupling module to isolate signals by order and an Order-wise Denoising module to denoise each order, coupled with a revised propagation scheme that prevents cross-order interference. Through experiments on three real-world datasets with synthetic noise, DRCSD consistently outperforms strong baselines and shows superior robustness to unstable interactions. The approach enables node-level personalized denoising and demonstrates practical potential for improving robustness and accuracy in noisy recommendation settings.

Abstract

Although the collaborative filtering (CF) algorithm has achieved remarkable performance in recommendation systems, it suffers from suboptimal recommendation performance due to noise in the user-item interaction matrix. Numerous noise-removal studies have improved recommendation models, but most existing approaches conduct denoising on a single graph. This may cause attenuation of collaborative signals: removing edges between two nodes can interrupt paths between other nodes, weakening path-dependent collaborative information. To address these limitations, this study proposes a novel GNN-based CF model called DRCSD for denoising unstable interactions. DRCSD includes two core modules: a collaborative signal decoupling module (decomposes signals into distinct orders by structural characteristics) and an order-wise denoising module (performs targeted denoising on each order). Additionally, the information aggregation mechanism of traditional GNN-based CF models is modified to avoid cross-order signal interference until the final pooling operation. Extensive experiments on three public real-world datasets show that DRCSD has superior robustness against unstable interactions and achieves statistically significant performance improvements in recommendation accuracy metrics compared to state-of-the-art baseline models.

Denoised Recommendation Model with Collaborative Signal Decoupling

TL;DR

The paper tackles noise in user-item interaction graphs that degrades collaborative filtering performance. It introduces DRCSD, a GNN-based framework with a Collaborative Signal Decoupling module to isolate signals by order and an Order-wise Denoising module to denoise each order, coupled with a revised propagation scheme that prevents cross-order interference. Through experiments on three real-world datasets with synthetic noise, DRCSD consistently outperforms strong baselines and shows superior robustness to unstable interactions. The approach enables node-level personalized denoising and demonstrates practical potential for improving robustness and accuracy in noisy recommendation settings.

Abstract

Although the collaborative filtering (CF) algorithm has achieved remarkable performance in recommendation systems, it suffers from suboptimal recommendation performance due to noise in the user-item interaction matrix. Numerous noise-removal studies have improved recommendation models, but most existing approaches conduct denoising on a single graph. This may cause attenuation of collaborative signals: removing edges between two nodes can interrupt paths between other nodes, weakening path-dependent collaborative information. To address these limitations, this study proposes a novel GNN-based CF model called DRCSD for denoising unstable interactions. DRCSD includes two core modules: a collaborative signal decoupling module (decomposes signals into distinct orders by structural characteristics) and an order-wise denoising module (performs targeted denoising on each order). Additionally, the information aggregation mechanism of traditional GNN-based CF models is modified to avoid cross-order signal interference until the final pooling operation. Extensive experiments on three public real-world datasets show that DRCSD has superior robustness against unstable interactions and achieves statistically significant performance improvements in recommendation accuracy metrics compared to state-of-the-art baseline models.

Paper Structure

This paper contains 26 sections, 1 theorem, 17 equations, 5 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. Overview
  4. Signal Decoupling Module
  5. Order-wise Denoised Module
  6. New Message Propagation
  7. Model Training
  8. Experiment
  9. Experimental Setup
  10. Dataset
  11. Baseline Models
  12. Implementation Details
  13. OverPerformance
  14. Further Analysis
  15. Hyperparameter study
  16. ...and 11 more sections

Key Result

Theorem 1

Let $G=(V,E)$ be a graph, where $V$ is the set of vertices and $E$ is the set of edges. For any two vertices $u,v \in V$, define $dG(u,v)$ as the length of the shortest path between $u$ and $v$ in $G$. For any positive integer $l$, define $G^l$ as the $l$-th power of $G$, in which two vertices $u$ a

Figures (5)

  • Figure 1: One possible situation: Since the interaction between $u_1$ and $i_1$ is judged as a noisy interaction, the path between $u_3$ and $u_1$ is cut off, resulting in the weakening of the cooperative signal aggregated by $u_3$.
  • Figure 2: Overview of DRCSD model, except classical GNN module, DRCSD model also contains the signal decoupling module, the order-wise denoised module. Signal decoupling module decouples each order collaborative into different interaction grpah, and oder-wise denoised module removes the noise in the signals of each order. In addition, we modify the classical GNN propagation mechanism to prevent the mixing of signals of all orders during propagation.
  • Figure 3: The running results of the DRCSD, SGL and RGCF models on three datasets with gradually increasing noise from 5% to 20%
  • Figure 4: DRCSD performance of different layers L
  • Figure 5: DRCSD performance of different denoised coefficient $\beta$

Theorems & Definitions (1)

  • Theorem 1