AI Agent-Driven Framework for Automated Product Knowledge Graph Construction
AI Agent-Driven Framework for Automated Product Knowledge Graph Construction in E-Commerce[edit]
Overview[edit]
This paper by Dimitar Peshevski, Riste Stojanov, and Dimitar Trajanov (2025) introduces a fully automated, AI agent-driven framework for constructing product knowledge graphs directly from unstructured product descriptions in e-commerce.
Problem Statement[edit]
The rapid expansion of e-commerce platforms generates vast amounts of unstructured product data, creating significant challenges for information retrieval, recommendation systems, and data analytics. While Knowledge Graphs (KGs) offer a structured, interpretable format to organize such data, constructing product-specific KGs remains a complex and largely manual process.
Approach[edit]
The framework leverages Large Language Models (LLMs) and operates in three stages using dedicated agents:
- Ontology Creation and Expansion — Agents generate and expand the product ontology from raw descriptions
- Ontology Refinement — Agents refine the ontology to ensure semantic coherence and minimize redundancy
- Knowledge Graph Population — Agents populate the final KG with structured knowledge extracted from unstructured data
This agent-based approach ensures semantic coherence, scalability, and high-quality output without relying on predefined schemas or handcrafted extraction rules.
Results[edit]
Evaluated on a real-world dataset of air conditioner product descriptions:
- Over 97% property coverage
- Minimal redundancy
- No predefined schemas or handcrafted extraction rules needed
Significance[edit]
Highlights the potential of LLMs to automate structured knowledge extraction in retail, providing a scalable path toward intelligent product data integration and utilization.
Relevance to Lou's Research[edit]
This paper is directly relevant to the discussion about agents performing semantic search over knowledge graphs. It demonstrates a practical implementation of LLM-driven agents autonomously building KGs from unstructured data — aligning with the vision of using agents for array semantic search, recursive inference, and runtime knowledge synthesis.
