Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

C$^3$: Capturing Consensus with Contrastive Learning in Group Recommendation

Soyoung Kim, Dongjun Lee, Jaekwang Kim

TL;DR

This work proposes Capturing Consensus with Contrastive Learning in Group Recommendation (C$^3), which focuses on exploring the consensus behind group decision-making, and significantly outperforms state-of-the-art baselines in both user and group recommendation tasks.

Abstract

Group recommendation aims to recommend tailored items to groups of users, where the key challenge is modeling a consensus that reflects member preferences. Although several existing deep learning models have achieved performance improvements, they still fail to capture consensus in various aspects: (1) Capturing consensus in small-group (2~5 members) recommendation systems, which align more closely with real-world scenarios, remains a significant challenge; (2) Most existing models significantly enhance the overall group performance but struggle with balancing individual and group performance. To address these issues, we propose Capturing Consensus with Contrastive Learning in Group Recommendation (C$^3$), which focuses on exploring the consensus behind group decision-making. A Transformer encoder is used to learn both group and user representations, and contrastive learning mitigates overfitting for users with many interactions, yielding more robust group representations. Experiments on four public datasets demonstrate that C$^3$ significantly outperforms state-of-the-art baselines in both user and group recommendation tasks.

C$^3$: Capturing Consensus with Contrastive Learning in Group Recommendation

TL;DR

This work proposes Capturing Consensus with Contrastive Learning in Group Recommendation (C$^3), which focuses on exploring the consensus behind group decision-making, and significantly outperforms state-of-the-art baselines in both user and group recommendation tasks.

Abstract

Group recommendation aims to recommend tailored items to groups of users, where the key challenge is modeling a consensus that reflects member preferences. Although several existing deep learning models have achieved performance improvements, they still fail to capture consensus in various aspects: (1) Capturing consensus in small-group (2~5 members) recommendation systems, which align more closely with real-world scenarios, remains a significant challenge; (2) Most existing models significantly enhance the overall group performance but struggle with balancing individual and group performance. To address these issues, we propose Capturing Consensus with Contrastive Learning in Group Recommendation (C), which focuses on exploring the consensus behind group decision-making. A Transformer encoder is used to learn both group and user representations, and contrastive learning mitigates overfitting for users with many interactions, yielding more robust group representations. Experiments on four public datasets demonstrate that C significantly outperforms state-of-the-art baselines in both user and group recommendation tasks.

Paper Structure

This paper contains 22 sections, 12 equations, 4 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Method
  3. Task Definition
  4. C3 Framework
  5. Embedding Layer.
  6. User & Group Encoder.
  7. Prediction Layer.
  8. Optimization for Recommendation Task
  9. Contrastive Learning
  10. Augmentation.
  11. Self-supervised Learning Signal.
  12. Experiments
  13. Experimental Setup
  14. Datasets.
  15. Evaluation Metrics.
  16. ...and 7 more sections

Figures (4)

  • Figure 1: Comparison of group recommendation scenarios a red-dotted example where one dominant user shifts the choice to Europe, and a blue-dotted example where full preference aggregation yields an Asia-leaning consensus.
  • Figure 2: Overview of our proposed C3, consisting of the transformer encoder, margin loss, and contrastive loss.
  • Figure 3: Visualization of the original representations (blue dots) and masked representations (red dots), where 80% of random users in a group are removed on Mafengwo.
  • Figure 4: NDCG@K comparison across group sizes on the Mafengwo dataset.