This article outlines a new Natural Language Processing (NLP) pipeline designed for constructing knowledge graphs (KGs) from unstructured scientific texts. The pipeline uniquely fine-tunes large language models (LLMs) using a small dataset to extract structured information, significantly enhancing the authenticity and traceability of the results. The case study involves the development of the Functional Material Knowledge Graph (FMKG) based on abstracts from 150,000 peer-reviewed papers, demonstrating its effectiveness and potential applications across various research dimensions.
Our study introduces a novel NLP pipeline for knowledge graph construction, capable of fine-tuning LLMs with minimal data to extract structured info from extensive unstructured texts.
The process maximizes the authenticity and traceability of structured information, allowing for the construction of a Functional Material Knowledge Graph (FMKG) from 150,000 peer-reviewed papers.
#natural-language-processing #knowledge-graph #scientific-research #machine-learning #data-extraction
Collection
[
|
...
]