Silo to Structure – A Pipeline for Extracting Knowledge Graphs from Historical Sources with LLMs

We introduce a computational pipeline that transforms heterogeneous historical sources into knowledge graphs using large language models (LLMs). It combines LLM-based open information extraction (OIE) with competency-question-guided ontology building and is grounded in three principles: task decomposition, human-in-the-loop validation, and research-question- and data-driven design.