.png){kind=icon}
.png){kind=icon}
Return
Multi-representation space recommendation with graph contrastive learning
.png){kind=icon}
PRE
AI
DOI：10.1016/j.eswa.2025.129274.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
• We propose a novel recommendation algorithm, MRSGCL, that integrates multi-representation space learning with graph contrastive learning to address the data sparsity challenge in collaborative filtering. • MRSGCL enhances recommendation performance by learning fine-grained user-item preferences from both heterogeneous and homogeneous graph structures. • The adaptive dual-representation space feature encoding module extracts embeddings for users and items in distinct latent spaces, improving preference representation. • The interest space alignment module uses contrastive learning to unify interest distributions across the two spaces while preserving the unique information from each. • We demonstrate the effectiveness and robustness of MRSGCL through extensive experiments on three public datasets, achieving state-of-the-art performance and proving the model’s scalability and high applicability.
Keywords:
multi-representation space learning
graph contrastive learning
collaborative filtering
user-item preferences
adaptive feature encoding
Journal
IF:
7.5
Papers:
2.9W
Citations:
10.2W