Mitigating Pattern Over-Generalization in Knowledge Graph Embeddings

The PogRE method addresses pattern over-generalization in Knowledge Graph Embeddings by using dense linear transformations and compound operations. This prevents models from universally applying learned inference patterns like symmetry or inversion to all related instances.