Virtual Neighbor

Virtual neighbor (VN) methods address the challenge of efficiently incorporating newly emerging entities into knowledge graph embeddings. Current research focuses on generating "virtual neighbors" for these entities using rule-based inference and graph neural networks, often incorporating iterative learning to refine both the embeddings and the virtual neighbor relationships. This approach improves the accuracy and robustness of knowledge graph completion tasks, particularly when dealing with entities having sparse connections, thereby enhancing the performance of various knowledge-based applications.