Tab2Onto: Unsupervised Semantification with Knowledge Graph Embeddings

Hamada M. Zahera, Stefan Heindorf, Stefan Balke, Jonas Haupt, Martin Voigt, Carolin Walter, Fabian Witter, Axel-Cyrille Ngonga Ngomo

Abstract: A large amount of data is generated every day by different systems and applications. In many cases, this data comes in a tabular format that lacks semantic representation and poses new challenges in data modelling. For semantic applications, it then becomes necessary to lift the data to a richer representation, such as a knowledge graph that adheres to a semantic ontology. We propose Tab2Onto, an unsupervised approach for learning ontologies from tabular data using knowledge graph embeddings, clustering, and a human in the loop. We conduct a set of experiments to investigate our approach on a benchmarking dataset from a medical domain and learn the ontology of diseases. Our code and datasets are provided at https://tab2onto.dice-research.org/ .