Using Large Language Models for searching explainable relations in a cloud of Cultural Heritage knowledge graphs

Knowledge discovery of “interesting” or even serendipitous relations in data, often called relational search, provides a novel Artificial Intelligence-based approach in Digital Humanities for studying Cultural Heritage. Relational search methods are traditionally symbolic, based on searching connections in knowledge graphs. In contrast, this paper presents a novel neuro-symbolic approach to relational search based on combining Large Language Models (LLM) with knowledge graphs (KG). It is argued that by using curated KG data and data models with Retrieval-Augmented Generation, hallucinations of LLMs can be mitigated and relational search extendedalso to web resources external to the underlying cloud of KGs. As a practical use case, first results of using the method for knowledge discovery as part of the new web service SampoSampo – Connecting Everything to Everything Else are presented.