Daily Arxiv

This page organizes papers related to artificial intelligence published around the world.
This page is summarized using Google Gemini and is operated on a non-profit basis.
The copyright of the paper belongs to the author and the relevant institution. When sharing, simply cite the source.

Efficient and Transferable Agentic Knowledge Graph RAG via Reinforcement Learning

Created by
  • Haebom

Author

Jinyeop Song, Song Wang, Julian Shun, Yada Zhu

Outline

This paper introduces KG-R1, a novel framework for augmented generation (RAG) of knowledge graphs (KGs). KG-R1 utilizes reinforcement learning (RL) to allow a single agent to interact with a KG, retrieving information at each step and incorporating it into inference and generation. This process is optimized through end-to-end RL. On the KGQA benchmark, KG-R1 demonstrates efficiency and transferability, achieving higher accuracy than existing methods despite a smaller model size using the Qwen-2.5-3B model. Furthermore, after training, KG-R1 can be applied to new KGs without modification.

Takeaways, Limitations

•
Takeaways:
◦
Reduced inference costs and specific KG dependencies through a single-agent architecture.
◦
Excellent performance even with smaller models (using Qwen-2.5-3B).
◦
High transferability to new KG (plug and play).
◦
Presentation of the KG-RAG framework to increase practical applicability.
•
Limitations:
◦
The specific Limitations is not explicitly mentioned in the paper.
◦
The complexity and training resource consumption of reinforcement learning-based models.
◦
Performance was only validated for a specific benchmark (KGQA). Further analysis of generalization performance for other question types and KGs is needed.
👍
Made with Slashpage