Semantic information-based attention mapping network for few-shot knowledge graph completion

Abstract

Few-shot Knowledge Graph Completion (FKGC), an emerging technology capable of inferring new triples using only a few reference relation triples, has gained significant attention in recent years. However, existing FKGC methods primarily focus on structural information while failing to effectively utilize the textual semantic information inherent in triples. To address this limitation, we propose an innovative Semantic Information-based Attention Mapping Network (SI-AMN). This novel model significantly enhances knowledge graph completion accuracy through a unique dual-information fusion mechanism that effectively integrates both structural and textual semantic information. The core innovation of SI-AMN lies in its two key components: a semantic encoder for extracting high-quality textual features and an attention mapping network that learns semantic interactions between entity and relation types. Experimental results on benchmark datasets demonstrate SI-AMN’s superior performance, achieving a 40% improvement in prediction accuracy compared to state-of-the-art methods. Ablation studies further validate the effectiveness of each component in our proposed model. This research not only provides a novel solution for knowledge graph completion but also reveals the crucial value of semantic information in graph completion tasks, paving the way for future research directions in this field.