Pretrain-KGE: Learning Knowledge Representation from Pretrained Language Models

Abstract

Conventional knowledge graph embedding (KGE) often suffers from limited knowledge representation, leading to performance degradation especially on the low-resource problem.