LLM-Powered Knowledge Graphs vs Topic Modelling for Sustainable Development Analytics

<p>Knowledge Graphs (KGs) are increasingly recognised as effective tools for structuring complex sustainability information and supporting Environmental, Social, and Governance (ESG)-aligned decision-making. This paper presents a comparative study between two approaches applied to sustainability analytics: (1) topic extraction using Generative Artificial Intelligence (GenAI), and (2) GenAI-driven KG construction using Neo4j’s Large Language Model (LLM) Graph Builder (GB). We apply both methods to a hybrid dataset comprising International Monetary Fund (IMF) Global Financial Stability Reports (GFSRs) and Greenstone ESG newsletters. While prior studies have leveraged topic modelling to identify thematic patterns in alignment with the United Nations (UN) Sustainable Development Goals (SDGs), our approach extends this by generating semantic graphs that explicitly capture entities, relationships, and development themes. Our findings demonstrate that GenAI-based KG construction provides more structured, interpretable, and actionable representations than topic modelling alone. These results highlight the potential of LLM-powered KGs to enhance transparency, insight generation, and alignment with SDGs—particularly in underrepresented domains such as the social and governance pillars of ESG.</p>