Large Models and Multimodal: A Survey of Cutting-Edge Approaches to Knowledge Graph Completion

The critical task of knowledge graph completion (KGC) cannot be overlooked when it comes to the evolution and application of new-generation knowledge graphs. With the advancement of multimodal learning and the rise of large models, knowledge graphs have experienced unprecedented development. Researchers have significantly enhanced KGC by integrating these emerging technologies with knowledge graphs. Within this context, this paper systematically reviews the evolutionary process of KGC methods, ranging from traditional representation learning approaches to those based on pre-training models, large language models (LLMs), and multimodal techniques. Specifically, we outline the application and efficacy of these emerging methods in addressing KGC problems, emphasizing their strengths and limitations in understanding knowledge associations and handling complex semantic information. Finally, we outline future developmental directions, aiming to portray a comprehensive and in-depth perspective of KGC from the latest vantage point for researchers, offering valuable insights for future research and applications.