NSE-KGC: A knowledge graph completion method based on neighborhood semantic evidence

The objective of knowledge graph completion (KGC) is to enhance our understanding of known facts and infer missing relationships within the graph. Previous text-based methods have advanced the field by incorporating effective contrastive learning techniques and improving learning efficiency through the construction of three distinct types of negative samples. Despite these advancements, these methods do not leverage the structural information inherent in the knowledge graph. In this paper, we introduce NSE-KGC, a novel model for knowledge graph completion that capitalizes on random walk neighbor semantic evidence aggregation. NSE-KGC explicitly models neighbor patterns identified through random walks at each layer and integrates these patterns through multi-layer aggregation. Extensive experiments conducted on two benchmark datasets, FB15k-237 and WN18RR, demonstrate that NSE-KGC achieves state-of-the-art performance in knowledge graph completion and offers superior entity extrapolation capabilities.