Daily Arxiv

This is a page that curates AI-related papers published worldwide.
All content here is summarized using Google Gemini and operated on a non-profit basis.
Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.

JEPA4Rec: Learning Effective Language Representations for Sequential Recommendation via Joint Embedding Predictive Architecture

Created by
  • Haebom

Author

Minh-Anh Nguyen, Dung D. Le

Outline

JEPA4Rec is a novel framework for sequential recommendation systems, proposed to address the challenges of a lack of understanding of common user preferences and data insufficiency. It converts descriptive information, such as item titles and categories, into sentences and uses a bidirectional Transformer encoder to learn semantically rich and transferable representations. Using masking techniques and self-supervised learning, it learns generalized item embeddings and improves recommendation performance. It outperforms existing state-of-the-art methods on several real-world datasets, demonstrating its effectiveness in cross-domain, cross-platform, and low-resource environments.

Takeaways, Limitations

•
Takeaways:
◦
Presenting a novel approach that effectively addresses the problems of data scarcity and lack of understanding of general user preferences.
◦
Improving recommendation performance by learning semantically rich and transferable item representations.
◦
Efficient model learning and performance improvement through self-supervised learning.
◦
Excellent performance across domains, platforms, and low-resource environments.
•
Limitations:
◦
Potential increase in computational cost due to the complexity of the proposed model.
◦
Possibility of information loss during conversion to sentences.
◦
It may only perform well on certain types of data (further research on generalization performance is needed).
◦
Lack of detailed analysis of the characteristics of the dataset used.
👍
Made with Slashpage