We want to predict the type of a DBpedia resource from its structure in the Knowledge graph. our preliminary study concludes that we can achieve it with accuracy above 90%.
Paper submitted to ICWE 2018, as short paper.
Type information, which is useful for responding many queries, plays a key role in Semantic Web. Nevertheless, it is common that type information of some instances is not present in knowledge graphs. Thus, type prediction of a given instance using background knowledge is an important knowledge graph completion task. To this end, the objective of this paper is to propose a data-driven type prediction approach using the structural information of the given instance utilising machine learning techniques. The experiments presented in the paper demonstrate that a binary classifier with structural information as features can be effectively used for type prediction of RDF knowledge graphs with high accuracy. The accuracy of the classifier is related to the diversity of training data as well as the how conceptually similar the different classes in the training and test data. Further, the experiments demonstrate that it is possible to build universal classifiers to a given class, i.e., a model training on one dataset can produce good predictions for another dataset in cases where training data contains conceptually different classes. For example, a model training on the English DBpedia can be used to predict types of the Spanish DBpedia.