An Automated Approach for Domain-Specific Knowledge Graph Generation─Graph Measures and Characterization.

Area of Science:

• Computational Chemistry
• Natural Language Processing
• Data Science

Background:

• The exponential growth of scientific publications necessitates automated methods for information extraction.
• Existing knowledge graph extraction methods are often domain-specific and limited.
• Bridging the gap between AI advancements and scientific literature analysis is crucial.

Purpose of the Study:

• To develop a novel natural language processing (NLP) approach for extracting knowledge graphs from technical documents.
• To create a semantically structured network (SSN) from synthetic chemistry patents.
• To characterize the resulting knowledge graph for linguistic and trend analysis.

Main Methods:

• Developed a natural language processing (NLP) model for knowledge graph extraction.
• Applied the model to approximately 100,000 full-length synthetic chemistry patents.
• Performed graph characterization using network motif structures, assortativity, and eigenvector centrality.

Main Results:

• Successfully extracted a semantically structured network (SSN) from chemical patents.
• Identified linguistic patterns in chemical reaction discourse, including common solvents and compound naming.
• Observed power-law trends in larger text corpora, indicating scalability.

Conclusions:

• The developed NLP approach provides a robust method for knowledge graph extraction in specialized domains.
• Quantitative characterization of knowledge graphs aids in understanding scientific discourse and validating large datasets.
• This work facilitates deeper insights into chemical synthesis literature and enables cross-domain knowledge graph comparisons.