Scientific knowledge graph and ontology generation using open large language models

Area of Science:

• Scientific knowledge representation
• Artificial Intelligence in Chemistry
• Data Mining in Scientific Literature

Background:

• Knowledge Graphs (KGs) and ontologies are crucial for structured information and enhancing Large Language Models (LLMs).
• Scientific domains often lack suitable ontologies for complex data, and manual KG/ontology curation is labor-intensive.
• Existing methods struggle with the unstructured nature of data in rapidly evolving scientific fields.

Purpose of the Study:

• To propose a novel, zero-shot, end-to-end method for generating ontologies and KGs from scientific literature using open-source LLMs.
• To address the limitations of manual curation and the absence of comprehensive ontologies in specialized scientific domains.
• To demonstrate the feasibility of automated knowledge representation for scientific research and knowledge management.

Main Methods:

• Leveraging open-source Large Language Models (LLMs) for automated ontology and KG generation.
• Implementing a zero-shot, end-to-end approach requiring no pre-existing domain-specific ontologies.
• Evaluating the method's performance by reconstructing existing KGs/ontologies and applying it to the Single Atom Catalysts (SACs) field.

Main Results:

• Successfully reconstructed KG and ontology for chemical elements and functional groups.
• Demonstrated effectiveness in generating structured knowledge from scarce and unstructured data in the Single Atom Catalysts (SACs) domain.
• Validated the ability of LLMs to automate the creation of complex scientific knowledge representations.

Conclusions:

• The proposed LLM-based method effectively automates ontology and KG generation from scientific literature.
• This approach significantly reduces the time and expertise required for knowledge structuring in specialized scientific fields.
• The generated KGs and ontologies enhance information retrieval and reasoning, paving the way for advanced LLM-assisted scientific discovery.