Automatic Ontology Construction Using LLMs as an External Layer of Memory, Verification, and Planning for Hybrid Intelligent Systems
Summary: arXiv:2604.20795v1 Announce Type: new
This paper presents a hybrid architecture for intelligent systems in which large language models (LLMs) are extended with an external ontological memory layer. Instead of relying solely on parametric knowledge and vector-based retrieval (RAG), the proposed approach constructs and maintains a structured knowledge graph using RDF/OWL representations, enabling persistent, verifiable, and semantically grounded reasoning.
Core Contributions
The core contribution of this research is an automated pipeline designed for ontology construction from heterogeneous data sources, which include:
- Documents
- APIs
- Dialogue logs
This system executes several critical processes:
- Entity recognition
- Relation extraction
- Normalization
- Triple generation
Following these processes, validation is performed using SHACL and OWL constraints, enabling continuous graph updates. The inference phase sees LLMs working over a combined context that integrates vector-based retrieval with graph-based reasoning and external tool interaction.
Experimental Observations
Experimental observations on planning tasks, including the Tower of Hanoi benchmark, indicate that ontology augmentation significantly enhances performance in multi-step reasoning scenarios compared to baseline LLM systems. In addition, the ontology layer facilitates formal validation of generated outputs, effectively transforming the system into a generation-verification-correction pipeline.
Addressing Key Limitations
The proposed architecture seeks to address several key limitations that are prevalent in current LLM-based systems:
- Lack of long-term memory
- Weak structural understanding
- Limited reasoning capabilities
By integrating an external ontological layer, the system provides a robust foundation for the development of agent-based systems, robotics applications, and enterprise AI solutions. These applications require persistent knowledge, explainability, and reliable decision-making, which are essential for advancing the capabilities of intelligent systems.
Conclusion
The introduction of an external ontological memory layer in conjunction with LLMs marks a significant advancement in the field of artificial intelligence. This hybrid architecture not only enhances reasoning capabilities but also ensures that generated knowledge is verifiable and structured, paving the way for future developments in intelligent systems.