Spatio-Temporal Meta-Graph Learning for Recommendation on Heterogeneous Graphs
Spatio-temporal information plays a crucial role in recommendation systems, and effectively capturing the spatiotemporal characteristics of user-item interactions is essential to enhance the performance of such systems. Conventional recommendation systems primarily focus on capturing interaction fea...
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
IEEE
2025-01-01
|
| Series: | IEEE Access |
| Subjects: | |
| Online Access: | https://ieeexplore.ieee.org/document/11122482/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Spatio-temporal information plays a crucial role in recommendation systems, and effectively capturing the spatiotemporal characteristics of user-item interactions is essential to enhance the performance of such systems. Conventional recommendation systems primarily focus on capturing interaction features within the data, frequently neglecting the significant influence of this information. To address this issue, this paper proposes a framework that combines graph ODE and meta-graph (MG) search. The graph ODE effectively captures spatiotemporal features across different time steps, resulting in node features that are enriched with semantic information. In addition, the MG search diverges from the conventional method of predefining MGs using a neural architecture search technique to derive MGs directly from the network. It also employs an attention mechanism to recommend the most suitable MGs once they have been generated. The proposed framework effectively integrates the spatiotemporal information of the nodes into the features, thereby enhancing understanding of the users’ needs and improving recommendation accuracy. The proposed STMGL framework was compared against the FMRec, HeteRec, SemRec, and GraphRec models on the MovieLens, Amazon, Douban, and Yelp datasets. The experimental results demonstrate a significant improvement the proposed STMGL framework’s performance compared with these existing recommendation techniques. |
|---|---|
| ISSN: | 2169-3536 |