ModphasE: An Efficient and Balanced Knowledge Graph Embedding Model for Large-Scale Graphs

Abstract

Abstract—Knowledge graphs or Knowledge networks (KGs or KNs), structured as triples of entities and relations, have become pivotal in diverse domains. However, the challenge of incomplete knowledge graphs, characterized by missing valid and test triples, catalyzes the domain of knowledge graph completions (KGCs), alternatively known as connection prediction. Motivated by the achievements of word embeddings, knowledge graph embeddings (KGEs) seek to grasp the semantic and structural interrelationships within the graphs via low-dimensional representations.This paper investigates the application of matrix decomposition techniques in knowledge graph embeddings, demonstrating their potential for efficient and effective link prediction. We introduce ModphasE, a novel method that achieves competitive result in comprehending complex relational patterns and reveals state-of-the-art (SOTA) results.