Unsupervised Type Constraint Inference in Bilinear Knowledge Graph Completion Models

Knowledge graph completion (KGC) models aim to provide a feasible way of manipulating facts in knowledge graphs. Most KGC models do not consider type constraint in relations due to the scarcity of type information in training data. We proposed an unsupervised method for inferring type constraint based on existing bilinear KGC models. Our method induces two type indicators into every relation and adjusts the location of entity embeddings in feature space to match the type indicators. Our approach eliminates the external feature space for entity types and type constraints in relations and has a consistent feature space; therefore, it has fewer parameters than other methods. Experiments show that our methods can improve the performance of the base models and outperform other methods on datasets about general knowledge.