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.

Action is All You Need: Dual-Flow Generative Ranking Network for Recommendation

Created by
  • Haebom

Author

Hao Guo, Erpeng Xue, Lei Huang, Shichao Wang, Xiaolei Wang, Lei Wang, Jinpeng Wang, Sheng Chen

Outline

This paper proposes a dual-flow generative ranking network (DFGR) to improve the efficiency of deep learning recommendation models (DLRM). Existing DLRMs rely on manual feature engineering, leading to high complexity and low scalability. Meta's HSTU-based model addressed these issues, but suffers from reduced training and inference efficiency due to increased input sequence length. DFGR addresses this issue by separating user action sequences into real and fake flows and redefining the interaction between the two flows within the QKV module of the self-attention mechanism. Experimental results demonstrate that DFGR outperforms existing recommendation models, including DLRM, Meta's HSTU, DIN, DCN, DIEN, and DeepFM, demonstrating that it is an efficient next-generation generative ranking paradigm with an optimal parameter allocation strategy under computational resource constraints.

Takeaways, Limitations

•
Takeaways:
◦
We reduced our reliance on manual feature engineering and improved model scalability and efficiency through end-to-end learning.
◦
Effectively solved the long input sequence problem of Meta's HSTU model, which is Limitations.
◦
We achieved performance that outperformed existing state-of-the-art recommendation models on a variety of datasets.
◦
We present an optimal parameter allocation strategy that provides efficient performance even in environments with limited computational resources.
•
Limitations:
◦
Further research may be needed to explore the generalization performance of the proposed model.
◦
Long-term stability and maintainability in real industrial environments need to be evaluated.
◦
Performance gains may be limited for certain types of datasets.
👍
Made with Slashpage