Knowledge graphs represent structured collections of assertions about the world. These valuable resource have been used in a wide range of applications. Knowledge graph embedding models have been effectively applied to tackle various graph-related problems, including link prediction. Yet, predictions of knowledge graph embedding models are often at most post-hoc and locally explainable. In contrast, predictions of concept learning models in description logics are ante-hoc and globally explainable. Although most knowledge graphs evolve with the time, previous works mostly focused on static knowledge graphs. In this project group, we will design algorithms that bring the both world together, namely, leverage embedding models to learn description logic concepts from examples on dynamic knowledge graphs.
The PG will be available in PAUL.