Contrastive Learning Based on Modality Reflection View for Accurate Multimedia Recommendation 


Vol. 13,  No. 11, pp. 637-644, Nov.  2024
https://doi.org/10.3745/TKIPS.2024.13.11.637


PDF
  Abstract

Recently, contrastive learning-based multimedia recommender systems have been actively researched. They provide significantly improved recommendation accuracy compared to existing multimedia recommender systems by generating views for items based on their multimodal features and conducting contrastive learning using these views. Nevertheless, our paper claims that existing contrastive learning-based multimedia recommender systems overlook the importance of properly reflecting the modality features of items when generating their views, thereby resulting in limited improvements in recommendation accuracy. This claim is based on findings from existing multimedia recommender systems, demonstrating that properly reflecting an item’s modality features in its embedding contributes to improved recommendation accuracy. Thus, our paper proposes a novel multimedia recommendation framework that conducts contrastive learning using views (spec., modality reflection views) that can properly reflect the modality features of items. Via experiments using two real-world public datasets, our paper demonstrated that the proposed method outperforms state-of-the-art multimedia recommender systems in recommendation accuracy by up to 6.42%. This result shows the importance of performing contrastive learning by utilizing modality reflection views.

  Statistics
  Cite this article

[IEEE Style]

S. Ban, T. Kim, S. Kim, "Contrastive Learning Based on Modality Reflection View for Accurate Multimedia Recommendation," The Transactions of the Korea Information Processing Society, vol. 13, no. 11, pp. 637-644, 2024. DOI: https://doi.org/10.3745/TKIPS.2024.13.11.637.

[ACM Style]

Sohee Ban, Taeri Kim, and Sang-Wook Kim. 2024. Contrastive Learning Based on Modality Reflection View for Accurate Multimedia Recommendation. The Transactions of the Korea Information Processing Society, 13, 11, (2024), 637-644. DOI: https://doi.org/10.3745/TKIPS.2024.13.11.637.

Forms

Search
(IN TITLE, AUTHOR, ABSTRACT,KEYWORDS)

Advanced Search ▼

POPULAR KEYWORDS
(TOP 10 KEYWORDS)

Recent Publications
(LAST 3 YEARS)

Old Journals

Indexing

All publications of TKIPS is indexed in DOI, EBSCO, Google Scholar, Crossref, and CrossCheck.

TKIPS is also selected as the Journal for Accreditation by NRF (National Research Foundation of Korea).

Related Journals