Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Leveraging Member-Group Relations via Multi-View Graph Filtering for Effective Group Recommendation

Chae-Hyun Kim, Yoon-Ryung Choi, Jin-Duk Park, Won-Yong Shin

TL;DR

This work tackles the challenge of group recommendations without expensive training by introducing Group-GF, a training-free approach based on multi-view graph filtering. It constructs two augmented item similarity graphs and one unified graph from member-, group-, and member-group interactions, learns distinct polynomial filters for each, and aggregates them to predict group-level scores, with a theoretical linkage to smoothness regularization. Empirically, Group-GF achieves state-of-the-art accuracy across three benchmarks while delivering substantial speedups due to simple matrix operations and training-free design. The approach enables fast, scalable group recommendations that effectively capture complex member-group dynamics.

Abstract

Group recommendation aims at providing optimized recommendations tailored to diverse groups, enabling groups to enjoy appropriate items. On the other hand, most existing group recommendation methods are built upon deep neural network (DNN) architectures designed to capture the intricate relationships between member-level and group-level interactions. While these DNN-based approaches have proven their effectiveness, they require complex and expensive training procedures to incorporate group-level interactions in addition to member-level interactions. To overcome such limitations, we introduce Group-GF, a new approach for extremely fast recommendations of items to each group via multi-view graph filtering (GF) that offers a holistic view of complex member-group dynamics, without the need for costly model training. Specifically, in Group-GF, we first construct three item similarity graphs manifesting different viewpoints for GF. Then, we discover a distinct polynomial graph filter for each similarity graph and judiciously aggregate the three graph filters. Extensive experiments demonstrate the effectiveness of Group-GF in terms of significantly reducing runtime and achieving state-of-the-art recommendation accuracy.

Leveraging Member-Group Relations via Multi-View Graph Filtering for Effective Group Recommendation

TL;DR

This work tackles the challenge of group recommendations without expensive training by introducing Group-GF, a training-free approach based on multi-view graph filtering. It constructs two augmented item similarity graphs and one unified graph from member-, group-, and member-group interactions, learns distinct polynomial filters for each, and aggregates them to predict group-level scores, with a theoretical linkage to smoothness regularization. Empirically, Group-GF achieves state-of-the-art accuracy across three benchmarks while delivering substantial speedups due to simple matrix operations and training-free design. The approach enables fast, scalable group recommendations that effectively capture complex member-group dynamics.

Abstract

Group recommendation aims at providing optimized recommendations tailored to diverse groups, enabling groups to enjoy appropriate items. On the other hand, most existing group recommendation methods are built upon deep neural network (DNN) architectures designed to capture the intricate relationships between member-level and group-level interactions. While these DNN-based approaches have proven their effectiveness, they require complex and expensive training procedures to incorporate group-level interactions in addition to member-level interactions. To overcome such limitations, we introduce Group-GF, a new approach for extremely fast recommendations of items to each group via multi-view graph filtering (GF) that offers a holistic view of complex member-group dynamics, without the need for costly model training. Specifically, in Group-GF, we first construct three item similarity graphs manifesting different viewpoints for GF. Then, we discover a distinct polynomial graph filter for each similarity graph and judiciously aggregate the three graph filters. Extensive experiments demonstrate the effectiveness of Group-GF in terms of significantly reducing runtime and achieving state-of-the-art recommendation accuracy.

Paper Structure

This paper contains 14 sections, 1 theorem, 12 equations, 3 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Preliminary
  3. Graph Filtering
  4. Problem Definition
  5. Methodology
  6. Augmented Graph Construction
  7. Unified Graph Construction
  8. Filter Design
  9. Experimental Evaluation
  10. Experimental Settings
  11. Runtime Analysis
  12. Recommendation Accuracy
  13. Ablation Study
  14. Conclusions and Future Work

Key Result

theorem 1

Let $L_u = I - \bar{P}_u$, $L_g = I - \bar{P}_g$, and $L_{\text{uni}} = I - \bar{P}_{\text{uni}}$ be the graph Laplacians corresponding to $\bar{P}_u$, $\bar{P}_g$, and $\bar{P}_{\text{uni}}$, respectively. Then, the predicted group preference scores $\mathbf{s}_g$ in Eq. group-gf are an approximate where $\lambda > 0$ is a regularization parameter and $0<\alpha,\beta<1$ are balancing parameters.

Figures (3)

  • Figure 1: Examples of both group and member consumption patterns on TV and trip platforms.
  • Figure 2: The schematic overview of Group-GF.
  • Figure 3: Eigenvalue distributions of the three item similarity graphs on the CAMRa2011 dataset.

Theorems & Definitions (1)

  • theorem 1