HILDEGARD Human-in-the-Loop Data Extraction and Graphically Augmented Relation Discovery

This research paper presents HILDEGARD, an application conceived to guide a semi-expert user in the domain of cultural heritage data management toward the creation of a lightweight knowledge graph tailored for supporting Automatic Story Generation (ASG).For this purpose, a subset of CIDOC-CRM classes and properties is preliminarily selected to fit the domain of interest. The input is con-stituted by one or more seed-heritage objects selected from a knowledge base. In our case study, they are SPARQL-queried from a LinkedOpen Database for Italian Cultural Heritage. The shortest path algorithm is then run online on all couplets obtained by a combination ofthe Wikipedia entities from the selected entry-seeds descriptions. The retrieved entities are subsequently linked to their related DBpedia-or YAGO-entry in the chosen language, and the relationships among them are automatically retrieved. The proposed tool addresses differ-ent knowledge gaps and societal needs simultaneously, such as the lack of solutions tailored for narrative purposes in the cultural heritagedomain, that is, to be used in a scenario where objects belonging to the same room must be linked through a narrative, which shall not onlybe coherent and informative but also engaging and interesting. The prototype, already able to generate the triples required for the follow-ing step of the proposed general ASG pipeline, is intended to be graphically enhanced so that the end user may guide the graph expansion interactively.