Topological Data Analysis and Graph-Based Learning for Multimodal Recommendation
Multimodal recommendation systems are becoming increasingly vital for delivering personalized content by utilizing various data sources, including text, images, and user interaction histories. However, current multimodal methods face challenges such as modality heterogeneity, data sparsity, and feat...
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
IEEE
2025-01-01
|
| Series: | IEEE Access |
| Subjects: | |
| Online Access: | https://ieeexplore.ieee.org/document/11048491/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1839648906755964928 |
|---|---|
| author | Khalil Bachiri Ali Yahyaouy Maria Malek Nicoleta Rogovschi |
| author_facet | Khalil Bachiri Ali Yahyaouy Maria Malek Nicoleta Rogovschi |
| author_sort | Khalil Bachiri |
| collection | DOAJ |
| description | Multimodal recommendation systems are becoming increasingly vital for delivering personalized content by utilizing various data sources, including text, images, and user interaction histories. However, current multimodal methods face challenges such as modality heterogeneity, data sparsity, and feature redundancy, which can result in less effective performance when dealing with complex, high-dimensional datasets. In this study, we present a new framework that combines Topological Data Analysis (TDA) with graph-based learning to improve multimodal recommendations (TDA-MMRec). Our approach captures higher-order dependencies and global structural patterns in multimodal data, enhancing the robustness and expressiveness of the representations we learn. By using persistent homology, we extract topological descriptors that convey stable structural information across different modalities, addressing the issues of sparsity and redundancy. We also introduce a modality-aware strategy for constructing graphs, which integrates features derived from TDA into multimodal similarity graphs to maintain both local and global structural properties. Furthermore, we propose a topological pruning technique that refines graph structures by removing redundant connections while preserving essential topological information, enhancing computational efficiency. Extensive experiments on large-scale multimodal datasets indicate that our TDA-augmented framework significantly outperforms leading multimodal recommendation models on key ranking metrics, including Precision@20, Recall@20, and NDCG@20. Our ablation studies confirm that topological descriptors are essential in boosting representation learning, especially in cold-start scenarios where traditional methods struggle due to data sparsity. |
| format | Article |
| id | doaj-art-c4827fb651824f26bb6bbdc0cb906b4b |
| institution | Matheson Library |
| issn | 2169-3536 |
| language | English |
| publishDate | 2025-01-01 |
| publisher | IEEE |
| record_format | Article |
| series | IEEE Access |
| spelling | doaj-art-c4827fb651824f26bb6bbdc0cb906b4b2025-06-27T23:00:56ZengIEEEIEEE Access2169-35362025-01-011310893410895410.1109/ACCESS.2025.358248011048491Topological Data Analysis and Graph-Based Learning for Multimodal RecommendationKhalil Bachiri0https://orcid.org/0000-0003-2500-7738Ali Yahyaouy1Maria Malek2Nicoleta Rogovschi3ETIS Laboratory, ENSEA, UMR8051, CNRS, CY Cergy Paris University, Cergy-Pontoise, FranceL3IA Laboratory, Faculty of Sciences Dhar El Mahraz, Sidi Mohamed Ben Abdellah University, Fez, MoroccoETIS Laboratory, ENSEA, UMR8051, CNRS, CY Cergy Paris University, Cergy-Pontoise, FranceLIPADE Laboratory, Université Paris Cité, Paris, FranceMultimodal recommendation systems are becoming increasingly vital for delivering personalized content by utilizing various data sources, including text, images, and user interaction histories. However, current multimodal methods face challenges such as modality heterogeneity, data sparsity, and feature redundancy, which can result in less effective performance when dealing with complex, high-dimensional datasets. In this study, we present a new framework that combines Topological Data Analysis (TDA) with graph-based learning to improve multimodal recommendations (TDA-MMRec). Our approach captures higher-order dependencies and global structural patterns in multimodal data, enhancing the robustness and expressiveness of the representations we learn. By using persistent homology, we extract topological descriptors that convey stable structural information across different modalities, addressing the issues of sparsity and redundancy. We also introduce a modality-aware strategy for constructing graphs, which integrates features derived from TDA into multimodal similarity graphs to maintain both local and global structural properties. Furthermore, we propose a topological pruning technique that refines graph structures by removing redundant connections while preserving essential topological information, enhancing computational efficiency. Extensive experiments on large-scale multimodal datasets indicate that our TDA-augmented framework significantly outperforms leading multimodal recommendation models on key ranking metrics, including Precision@20, Recall@20, and NDCG@20. Our ablation studies confirm that topological descriptors are essential in boosting representation learning, especially in cold-start scenarios where traditional methods struggle due to data sparsity.https://ieeexplore.ieee.org/document/11048491/Multimodal recommendation systemstopological data analysis (TDA)persistent homologygraph neural networks (GNNs)representation learningmultimodal fusion |
| spellingShingle | Khalil Bachiri Ali Yahyaouy Maria Malek Nicoleta Rogovschi Topological Data Analysis and Graph-Based Learning for Multimodal Recommendation IEEE Access Multimodal recommendation systems topological data analysis (TDA) persistent homology graph neural networks (GNNs) representation learning multimodal fusion |
| title | Topological Data Analysis and Graph-Based Learning for Multimodal Recommendation |
| title_full | Topological Data Analysis and Graph-Based Learning for Multimodal Recommendation |
| title_fullStr | Topological Data Analysis and Graph-Based Learning for Multimodal Recommendation |
| title_full_unstemmed | Topological Data Analysis and Graph-Based Learning for Multimodal Recommendation |
| title_short | Topological Data Analysis and Graph-Based Learning for Multimodal Recommendation |
| title_sort | topological data analysis and graph based learning for multimodal recommendation |
| topic | Multimodal recommendation systems topological data analysis (TDA) persistent homology graph neural networks (GNNs) representation learning multimodal fusion |
| url | https://ieeexplore.ieee.org/document/11048491/ |
| work_keys_str_mv | AT khalilbachiri topologicaldataanalysisandgraphbasedlearningformultimodalrecommendation AT aliyahyaouy topologicaldataanalysisandgraphbasedlearningformultimodalrecommendation AT mariamalek topologicaldataanalysisandgraphbasedlearningformultimodalrecommendation AT nicoletarogovschi topologicaldataanalysisandgraphbasedlearningformultimodalrecommendation |