Stochastic Deep Graph Clustering for Practical Group Formation

TL;DR

DeepForm tackles practical group formation in group recommender systems by jointly modeling high-order user relations, enabling real-time group reconfiguration, and supporting dynamic numbers of groups $K$. It uses a lightweight propagation-only GCN to encode high-order signals from a user–item rating graph, augmented with an autoencoder to incorporate item-context, and introduces stochastic cluster learning to allow inference-time adjustment of $K$ without retraining. A cluster-contrastive objective (triplet and InfoNCE losses) stabilizes embeddings across varying $K$, ensuring coherent and separable group structures. Across multiple datasets, DeepForm achieves superior group formation quality, faster group formation, and improved recommendation accuracy compared with strong baselines, highlighting its practicality for dynamic real-world GRS deployments.

Abstract

While prior work on group recommender systems (GRSs) has primarily focused on improving recommendation accuracy, most approaches assume static or predefined groups, making them unsuitable for dynamic, real-world scenarios. We reframe group formation as a core challenge in GRSs and propose DeepForm (Stochastic Deep Graph Clustering for Practical Group Formation), a framework designed to meet three key operational requirements: (1) the incorporation of high-order user information, (2) real-time group formation, and (3) dynamic adjustment of the number of groups. DeepForm employs a lightweight GCN architecture that effectively captures high-order structural signals. Stochastic cluster learning enables adaptive group reconfiguration without retraining, while contrastive learning refines groups under dynamic conditions. Experiments on multiple datasets demonstrate that DeepForm achieves superior group formation quality, efficiency, and recommendation accuracy compared with various baselines.

Figures (3)

• Figure 1: DeepForm Overview. High-order user representations are obtained from the rating matrix and user graph. Representations are refined in HTML]F4F9FDtraining via stochastic clustering and contrastive learning and clustered for any desired $K$ at HTML]FDEDEEinference.
• Figure 2: Efficiency comparison of deep graph clustering methods across various numbers of groups.
• Figure 3: Performance comparison across various numbers of groups for representative group formation methods.