정확한 멀티미디어 추천을 위한모달리티 반영 뷰 기반의 대조 학습
Contrastive Learning Based on Modality Reflection View for Accurate Multimedia Recommendation
0
조회
0
다운로드
0
인용
0
참고문헌
키워드
초록
최근, 대조 학습 기반의 멀티미디어 추천 시스템들이 활발하게 연구되고 있다. 이들은 아이템의 멀티모달 피처들로부터 아이템의 뷰들을 생성하고, 이러한 뷰들을 활용하여 대조 학습을 진행함으로써 기존 멀티미디어 추천 시스템들보다 상당히 향상된 추천 정확도를 제공한다. 그럼에도불구하고, 본 논문에서는 기존 대조 학습 기반의 멀티미디어 추천 시스템들이 아이템의 뷰들을 생성하는 데에 아이템의 모달리티 피처들을 올바르게반영하는 것의 중요성을 간과하며, 그 결과 추천 정확도 향상에 제약을 갖는다고 주장한다. 이는 아이템 임베딩에 아이템 자신의 모달리티 피처를올바르게 반영하는 것이 추천 정확도 향상에 도움이 된다는 기존 멀티미디어 추천 시스템들의 발견에 기반한다. 따라서 본 논문에서는 아이템의모달리티 피처들을 올바르게 반영하는 뷰들(구체적으로, 모달리티 반영 뷰들)을 활용하여 대조 학습을 진행하는 새로운 멀티미디어 추천 시스템을제안한다. 두 가지 실세계 공개 데이터 집합들에 대한 실험들을 통해, 본 논문은 제안 방안이 최신 멀티미디어 추천 시스템의 정확도를 6.42%까지개선할 수 있음을 확인하였으며, 이는 모달리티 반영 뷰들을 활용하여 대조 학습을 진행하는 것의 중요성을 뒷받침한다.
Abstract
Recently, contrastive learning-based multimedia recommender systems have been actively researched. They provide significantlyimproved recommendation accuracy compared to existing multimedia recommender systems by generating views for items based on theirmultimodal features and conducting contrastive learning using these views. Nevertheless, our paper claims that existing contrastivelearning-based multimedia recommender systems overlook the importance of properly reflecting the modality features of items whengenerating their views, thereby resulting in limited improvements in recommendation accuracy. This claim is based on findings fromexisting multimedia recommender systems, demonstrating that properly reflecting an item’s modality features in its embedding contributesto improved recommendation accuracy. Thus, our paper proposes a novel multimedia recommendation framework that conductscontrastive learning using views (spec., modality reflection views) that can properly reflect the modality features of items. Via experimentsusing two real-world public datasets, our paper demonstrated that the proposed method outperforms state-of-the-art multimediarecommender systems in recommendation accuracy by up to 6.42%. This result shows the importance of performing contrastive learningby utilizing modality reflection views.
참고문헌 (24)
- D. K. Chae, "Rating augmentation with generative adversarial networks towards accurate collaborative filtering", The World Wide Web Conference (2019)
- D. K. Chae, "AR-CF : Augmenting virtual users and items in collaborative filtering for addressing cold-start problems", Proceedings of the 43rd International ACM SIGIR Conference on Research and Development in Information Retrieval (2020)
- T. Y. Kong, "Linear, or non-linear, that is the question!", Proceedings of the Fifteenth ACM International Conference on Web Search and Data Mining (2022)
- H. J. Lim, "AiRS: a large-scale recommender system at naver news", 2022 IEEE 38th International Conference on Data Engineering (ICDE) (2022), IEEE
- Y. Su, "Neural graph matching based collaborative filtering", Proceedings of the 44th International ACM SIGIR Conference on Research and Development in Information Retrieval (2021)
- L. Chen, "Set2setRank : Collaborative set to set ranking for implicit feedback based recommendation", Proceedings of the 44th International ACM SIGIR Conference on Research and Development in Information Retrieval (2021)
- X. He, "Lightgcn : Simplifying and powering graph convolution network for recommendation", Proceedings of the 43rd International ACM SIGIR conference on research and development in Information Retrieval (2020)
- R. He, "VBPR: visual bayesian personalized ranking from implicit feedback", Proceedings of the AAAI Conference on Artificial Intelligence, 30(1) (2016)
- Y. Wei, "MMGCN : Multi-modal graph convolution network for personalized recommendation of micro-video", Proceedings of the 27th ACM international conference on multimedia (2019)
- Y. Wei, "Graphrefined convolutional network for multimedia recommendation with implicit feedback", Proceedings of the 28th ACM International Conference on Multimedia (2020)
- J. Zhang, "Mining latent structures for multimedia recommendation", Proceedings of the 29th ACM International Conference on Multimedia (2021)
- T. R. Kim, "MARIO : modality-aware attention and modality-preserving decoders for multimedia recommendation", Proceedings of the 31st ACM International Conference on Information &Knowledge Management (2022)
- KY. G. Kim, "MONET : Modality-Embracing Graph Convolutional Network and Target-Aware Attention for Multimedia Recommendation", Proceedings of the 17th ACM International Conference on Web Search and Data Mining (2024)
- A. Krizhevsky, "ImageNet Classification with Deep Convolutional Neural Networks", Proceedings of the Annual Conference on Neural Information Processing Systems (NeurIPS), pp.1106~1114 (2012)
- H. Sepp, "Long Short-Term Memory", Neural computation, pp.1735~1780 (1997)
- P. Yu, "Multi-view graph convolutional network for multimedia recommendation", Proceedings of the 31st ACM International Conference on Multimedia (2023)
- Z. Yi, "Multi-modal graph contrastive learning for micro-video recommendation", Proceedings of the 45th International ACM SIGIR Conference on Research and Development in Information Retrieval (2022)
- J. Zhang, "Latent structure mining with contrastive modality fusion for multimedia recommendation", IEEE Transactions on Knowledge and Data Engineering (2022)
- Z. Tao, "Self-supervised learning for multimedia recommendation", IEEE Transactions on Multimedia (2022)
- S. H. Ban, "Multimedia Recommender System Based on Contrastive Learning with Modality-Reflective View", Proceedings of the Annual Symposium of Korea Information Processing Society Conference (KIPS) 2024, 31(1), pp.635~638 (2024)
- J. McAuley, "Image-based recommendations on styles and substitutes", Proceedings of the 38th International ACM SIGIR Conference on Research and Development in Information Retrieval (2015)
- J. Wu, "Self-supervised graph learning for recommendation", Proceedings of the 44th international ACM SIGIR Conference on Research and Development in Information Retrieval (2021)