Search Results

Knowledge graph (KG) construction from heterogeneous data is a complex process that requires good understanding of the data to semantically annotate entities and their relations, and define the terms of the KG based on these entities and relations. Due to this complexity, KG construction remains primarily manual. On the one hand, semantic annotation systems focus on entity and relation disambiguation, but ultimately they do not construct a KG. On the other hand, declarative systems which typically construct the KGs, assume that the semantic annotations are already available. However, the two types of systems have not yet been effectively integrated. In this paper, we propose a formal method that integrates semantic annotation and declarative systems for automating end-to-end KG construction through automated declarative mappings generation. We validate our approach by creating AutoRML, a system that can leverage different semantic annotation frameworks to annotate tabular data and declaratively construct KGs. Evaluations demonstrate that AutoRML can construct KGs identical to manually constructed ones when the same target knowledge base is used as reference. AutoRML supports semi-automatic KG construction when semantic annotations are insufficient for full automation, generating human-friendly declarative mappings that can be refined by experts. We showcase AutoRML’s performance with diverse datasets, showing its potential in automating data integration workflows.