Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Consistency and Discrepancy-Based Contrastive Tripartite Graph Learning for Recommendations

Linxin Guo, Yaochen Zhu, Min Gao, Yinghui Tao, Junliang Yu, Chen Chen

TL;DR

This work tackles cold-start challenges in tripartite graph-based recommendations by introducing Consistency and Discrepancy-based graph contrastive learning (CDR). It defines reachable and non-reachable meta-paths to derive high-order consistency and discrepancy metrics, computable via the limit theory of GCN, and uses a contrastive CD loss to supervise embeddings without requiring direct interactions. The model features pre-training and fine-tuning variants (CDR-P and CDR-F) that fuse information from member interactions and tuple interactions, achieving superior performance on group and bundle recommendation tasks across Mafengwo, Youshu, and Last-FM, with notable robustness in extreme cold-start settings. Overall, CDR demonstrates the value of meta-path–driven supervision signals and a contrastive objective for effective tripartite graph learning, advancing practical recommendation systems in sparse data regimes.

Abstract

Tripartite graph-based recommender systems markedly diverge from traditional models by recommending unique combinations such as user groups and item bundles. Despite their effectiveness, these systems exacerbate the longstanding cold-start problem in traditional recommender systems, because any number of user groups or item bundles can be formed among users or items. To address this issue, we introduce a Consistency and Discrepancy-based graph contrastive learning method for tripartite graph-based Recommendation. This approach leverages two novel meta-path-based metrics consistency and discrepancy to capture nuanced, implicit associations between the recommended objects and the recommendees. These metrics, indicative of high-order similarities, can be efficiently calculated with infinite graph convolutional networks layers under a multi-objective optimization framework, using the limit theory of GCN.

Consistency and Discrepancy-Based Contrastive Tripartite Graph Learning for Recommendations

TL;DR

This work tackles cold-start challenges in tripartite graph-based recommendations by introducing Consistency and Discrepancy-based graph contrastive learning (CDR). It defines reachable and non-reachable meta-paths to derive high-order consistency and discrepancy metrics, computable via the limit theory of GCN, and uses a contrastive CD loss to supervise embeddings without requiring direct interactions. The model features pre-training and fine-tuning variants (CDR-P and CDR-F) that fuse information from member interactions and tuple interactions, achieving superior performance on group and bundle recommendation tasks across Mafengwo, Youshu, and Last-FM, with notable robustness in extreme cold-start settings. Overall, CDR demonstrates the value of meta-path–driven supervision signals and a contrastive objective for effective tripartite graph learning, advancing practical recommendation systems in sparse data regimes.

Abstract

Tripartite graph-based recommender systems markedly diverge from traditional models by recommending unique combinations such as user groups and item bundles. Despite their effectiveness, these systems exacerbate the longstanding cold-start problem in traditional recommender systems, because any number of user groups or item bundles can be formed among users or items. To address this issue, we introduce a Consistency and Discrepancy-based graph contrastive learning method for tripartite graph-based Recommendation. This approach leverages two novel meta-path-based metrics consistency and discrepancy to capture nuanced, implicit associations between the recommended objects and the recommendees. These metrics, indicative of high-order similarities, can be efficiently calculated with infinite graph convolutional networks layers under a multi-objective optimization framework, using the limit theory of GCN.
Paper Structure (42 sections, 22 equations, 6 figures, 9 tables, 1 algorithm)

This paper contains 42 sections, 22 equations, 6 figures, 9 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Preliminaries
  3. Relevant Concepts
  4. Problem Formulation
  5. Methodology
  6. Tuple-Object Metric Learning
  7. Optimization Objective
  8. Message Passing Mechanism
  9. Solution Based on the Limit Theory of GCN
  10. Solution with Reachable/Non-Reachable Meta-Paths
  11. Final Objective
  12. Consistency and Discrepancy Metrics
  13. Contrastive Learning-Inspired CD Loss
  14. Motivation of the CD Loss
  15. CD Loss
  16. ...and 27 more sections

Figures (6)

  • Figure 1: Comparison of bipartite and tripartite graph-based recommendation.
  • Figure 2: Two kinds of meta-paths, as examples, for deriving consistency and discrepancy metrics.
  • Figure 3: CDR Model Structure. Upper pre-training with tuple-member affiliations and member interactions, lower fine-tuning with tuple interactions, featuring variants CDR-P and CDR-F. CDR-P is for extreme cold-start scenes, whereas CDR-F is a variant that only uses tuple interactions.
  • Figure 4: Performance comparison of different loss functions of CDR on the Mafengwo dataset.
  • Figure 5: The effect of different temperatures $\tau$ on the pre-training and fine-tuning phases.
  • ...and 1 more figures

Theorems & Definitions (3)

  • Definition 1
  • Definition 2
  • Definition 3