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.

From Query to Logic: Ontology-Driven Multi-Hop Reasoning in LLMs

Created by
  • Haebom

Author

Haonan Bian, Yutao Qi, Rui Yang, Yuanxi Che, Jiaqian Wang, Heming Xia, Ranran Zhen

Outline

This paper presents ORACLE (Ontology-driven Reasoning and Chain for Logical Elucidation), a training-free framework that combines the structural advantages of knowledge graphs with the generative capabilities of LLMs to overcome the limitations of large-scale language models (LLMs) in complex multi-stage question answering (MQA) tasks. ORACLE consists of three steps: dynamically generating a question-specific knowledge ontology, transforming it into a first-order logical reasoning chain, and decomposing the original question into logically coherent subquestions. Experimental results on several MQA benchmarks demonstrate that ORACLE achieves competitive performance with state-of-the-art models such as DeepSeek-R1, demonstrating the effectiveness of each component and generating a more logical and interpretable reasoning chain than existing approaches.

Takeaways, Limitations

•
Takeaways:
◦
Presenting a new framework that contributes to improving LLM's multi-step question-answering ability.
◦
Combining the strengths of knowledge graphs and LLM to improve MQA performance.
◦
Generate logical and interpretable reasoning processes.
◦
No training required, reducing data dependency and increasing application flexibility.
◦
Achieve competitive performance with cutting-edge models.
•
Limitations:
◦
Further validation of the generalization performance of the proposed framework is needed.
◦
The possibility of dependencies on specific types of questions or knowledge graphs.
◦
Potential increase in computational cost due to the complexity of the ontology creation and inference process.
◦
Limited applicability to complex and ambiguous questions in the real world.
👍
Made with Slashpage