Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Dual Adversarial Perturbators Generate rich Views for Recommendation

Lijun Zhang, Yuan Yao, Haibo Ye

TL;DR

A dual-adversarial graph learning approach, AvoGCL, which emulates curriculum learning by progressively applying adversarial training to graph structures and embedding perturbations and significantly outperforms the state-of-the-art competitors.

Abstract

Graph contrastive learning (GCL) has been extensively studied and leveraged as a potent tool in recommender systems. Most existing GCL-based recommenders generate contrastive views by altering the graph structure or introducing perturbations to embedding. While these methods effectively enhance learning from sparse data, they risk performance degradation or even training collapse when the differences between contrastive views become too pronounced. To mitigate this issue, we employ curriculum learning to incrementally increase the disparity between contrastive views, enabling the model to gain from more challenging scenarios. In this paper, we propose a dual-adversarial graph learning approach, AvoGCL, which emulates curriculum learning by progressively applying adversarial training to graph structures and embedding perturbations. Specifically, AvoGCL construct contrastive views by reducing graph redundancy and generating adversarial perturbations in the embedding space, and achieve better results by gradually increasing the difficulty of contrastive views. Extensive experiments on three real-world datasets demonstrate that AvoGCL significantly outperforms the state-of-the-art competitors.

Dual Adversarial Perturbators Generate rich Views for Recommendation

TL;DR

A dual-adversarial graph learning approach, AvoGCL, which emulates curriculum learning by progressively applying adversarial training to graph structures and embedding perturbations and significantly outperforms the state-of-the-art competitors.

Abstract

Graph contrastive learning (GCL) has been extensively studied and leveraged as a potent tool in recommender systems. Most existing GCL-based recommenders generate contrastive views by altering the graph structure or introducing perturbations to embedding. While these methods effectively enhance learning from sparse data, they risk performance degradation or even training collapse when the differences between contrastive views become too pronounced. To mitigate this issue, we employ curriculum learning to incrementally increase the disparity between contrastive views, enabling the model to gain from more challenging scenarios. In this paper, we propose a dual-adversarial graph learning approach, AvoGCL, which emulates curriculum learning by progressively applying adversarial training to graph structures and embedding perturbations. Specifically, AvoGCL construct contrastive views by reducing graph redundancy and generating adversarial perturbations in the embedding space, and achieve better results by gradually increasing the difficulty of contrastive views. Extensive experiments on three real-world datasets demonstrate that AvoGCL significantly outperforms the state-of-the-art competitors.
Paper Structure (26 sections, 12 equations, 6 figures, 6 tables, 1 algorithm)

This paper contains 26 sections, 12 equations, 6 figures, 6 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Preliminaries
  3. Graph Collaborative Filtering
  4. Graph Contrastive Learning
  5. The Proposed Approach
  6. Adversarial Structure Perturbator
  7. Adversarial Embedding Perturbator
  8. Adversarial Training
  9. Complexity Analysis
  10. Experiments
  11. Experimental Setup
  12. Datasets
  13. Evaluation Metrics
  14. Compared Methods
  15. Hyper-parameter Settings
  16. ...and 11 more sections

Figures (6)

  • Figure 1: The curves of Recall@10 and NDCG@10 of the model with the edge deletion ratio. SGLC represents the SGL method with added curriculum learning. '-R' represents the Recall@10 score of the model, and '-N' represents the NDCG@10 score of the model.
  • Figure 2: The overview of AvoGCL.
  • Figure 3: The performance gain analysis of AvoGCL. Both the adversarial structure perturbator and the adversarial embedding perturbator can improve the performance of AvoGCL.
  • Figure 4: The performance on data with different sparsity degrees. The sparsity decreases from 'test1' to 'test5'. The relative improvement of AvoGCL in the sparsest case is even larger than the average.
  • Figure 5: Parameter sensitivity results of $\lambda_1$. The proposed AvoGCL performs relatively stable w.r.t. $\lambda_1$ in a wide range.
  • ...and 1 more figures