OntoCSM: Ontology-Aware Characteristic Set Merging for RDF Type Discovery

Pengkai Liu, Shunting Cai, Baozhu Liu, Xin Wang

Abstract: With the growing popularity and application of knowledge-based artificial intelligence, the scale of knowledge graph data is dramatically increasing. The RDF, as one of the mainstream models of knowledge graphs, is widely used to describe the characteristics of Web resources due to its simplicity and flexibility. However, RDF datasets are usually incomplete (without information) and noisy, which hinders downstream tasks. RDF entities can be characterized by their characteristic sets that is the sets of predicates of the RDF entities. Since untyped entities can be assigned to closest types by merging characteristic sets, optimally merging characteristic sets has become a crucial issue. In this paper, aiming at the Optimal Characteristic Set Merge Problem (OCSMP), we propose an Ontology-Aware Characteristic Set Merging algorithm, called OntoCSM, which extracts an ontology hierarchy using RDF characteristic sets and guides the merging process by optimizing the objective function. Extensive experiments on various datasets show that the efficiency of OntoCSM is generally higher than that of the state-of-the-art algorithms and can be improved by orders of magnitude in the best case. The accuracy and scalability of our method have been verified, which shows that OntoCSM can reach competitive results to the existing algorithms while being ontology-aware.