Knowledge graph (KG) link completion is a well-studied problem in AI with two major competing solution techniques: rule-based methods and embedding-based methods.
In this talk, we present a novel rule-based method for the KG link completion problem. Our method uses a relatively simple optimization formulation to choose
rules and assign weights to them. To foster interpretability of the chosen rules, we limit the complexity of the final collection of rules via explicit constraints. We show
that our method can obtain state-of-the-art results with relatively compact rule sets. This is joint work with Joao Goncalves and Francisco Barahona.
Back to Deep Learning and Combinatorial Optimization