# Customized Information and Domain-centric Knowledge Graph Construction with Large Language Models

**Authors:** Frank Wawrzik, Matthias Plaue, Savan Vekariya, Christoph Grimm  
**arXiv:** [2409.20010](https://arxiv.org/abs/2409.20010)  
**Date:** 2024-09-30  
**Summary page:** https://www.summarizepaper.com/en/paper/2409.20010v1/

## Key points

- Comprehensive framework for technology intelligence and planning using knowledge graphs in cyber-physical systems
- Methodology involves text mining process including information retrieval, keyphrase extraction, semantic network creation, and topic map visualization
- Querying a technology intelligence database with over 176 million news articles, patents, and research documents for relevant information extraction
- Knowledge graph construction pipeline utilizing tools like Owlready2, REBEL, and ChatGPT to create OWL Ontology Graph
- Reasoning procedures based on GENIAL! Basic Ontology schema to enhance consistency and structure of the resulting knowledge graph
- Innovation Graph database containing 50 million research publications, 81 million internet news articles, and 45 million patent filings from the past decade
- Resources provided include research articles graph on GitHub, keyphrases extraction, semantic networks, knowledge graphs in OWL and HTML formats created with ChatGPT
- Scalable approach for technology intelligence and planning in cyber-physical systems with significant improvements in class recognition compared to existing approaches

## Summary

, , , , 


In this paper, we present a comprehensive framework for technology intelligence and planning that leverages knowledge graphs to provide structured information for actionable insights in the realm of cyber-physical systems. The methodology involves a text mining process that encompasses information retrieval, keyphrase extraction, semantic network creation, and topic map visualization. By querying a technology intelligence database with over 176 million news articles, patents, and research documents, relevant information is extracted to facilitate data exploration. Keyphrases are utilized to create a semantic network which is visualized as a topic map, enabling interactive exploration of the domain. The highest scoring documents serve as input for the knowledge graph construction (KGC) pipeline. The KGC pipeline involves converting articles into .txt files which are transformed into an OWL Ontology Graph (.owl file) using tools like Owlready2 and transformer models such as REBEL and ChatGPT. The resulting knowledge graph is subjected to reasoning procedures based on the GENIAL! Basic Ontology schema to enhance consistency and structure. This approach ensures improved results in machine learning applications. The Innovation Graph database contains 50 million research publications, 81 million internet news articles, and 45 million patent filings from the past decade. Document genres from these sources are utilized to enrich the knowledge base of innovation and technology. Resources provided include the generated graph of research articles on GitHub along with extracted keyphrases, semantic networks, knowledge graphs in OWL and HTML formats created with ChatGPT. Additionally, an ontology for Innovation Knowledge Graph in OWL and HTML formats is available along with Wikidata queries related to the electronic domain. Overall, this framework offers a scalable approach for technology intelligence and planning in cyber-physical systems by harnessing knowledge graphs supported by advanced language models and reasoning procedures based on established ontologies. The methodology showcases significant improvements in class recognition, relationship construction, and categorization compared to existing approaches like GraphGPT or bi-LSTM models. Furthermore, it demonstrates adaptability across domains while emphasizing human oversight for content quality assurance at critical stages of development.