Integrating LLMs and the PrOntoNarr Ontology for Automated KG Construction in Multi-level Propaganda Narrative Analysis

Knowledge Graphs (KGs) are essential for advancing semantic web technologies, yet their construction from unstructured text, particularly in domains like political discourse, remains a significant challenge. Traditional rule-based and machine learning methods often fail to capture the structural and persuasive dimensions of underlying narratives. To address this gap, this study proposes a comprehensive framework for identifying, modeling, and querying narratives within propaganda across three levels: text, paragraph, and campaign. Central to this approach is the PrOntoNarr ontology, a formal semantic model that integrates narrative theory with propaganda analysis to facilitate the construction of a KG. Leveraging the generative capabilities of Large Language Models (LLMs), the framework automates KG construction by transforming unstructured propaganda texts into formal RDF triples through prompt engineering. The resulting KG provides a structured representation of narrative elements in political speeches and enables complex, multi-level analysis via SPARQL queries. The framework is validated using a dataset of political propaganda speeches from the 2025 Gaza ceasefire negotiations, demonstrating its capacity to support researchers in uncovering patterns and persuasive strategies employed in modern persuasive discourse.