S²DN: Learning to Denoise Unconvincing Knowledge for Inductive Knowledge Graph Completion

• DOI: 10.1609/aaai.v39i12.33346
• Publication Date: 2025-04-11T12:05:00Z
• Authors: Tengfei Ma; Yujie Chen; Liang Wang; Xuan Lin; Bosheng Song; Xiangxiang Zeng
• Keywords: Knowledge graph

Inductive Knowledge Graph Completion (KGC) aims to infer missing facts between newly emerged entities within knowledge graphs (KGs), posing a significant challenge. While recent studies have shown promising results in inferring such through subgraph reasoning, they suffer from (i) the semantic inconsistencies of similar relations, and (ii) noisy interactions inherent KGs due presence unconvincing for emerging entities. To address these challenges, we propose Semantic Structure-aware Denoising Network (S2DN) inductive KGC. Our goal is learn adaptable general semantics reliable structures distill consistent while preserving KGs. Specifically, introduce smoothing module over enclosing subgraphs retain universal relations. We incorporate structure refining filter out unreliable offer additional knowledge, retaining robust surrounding target links. Extensive experiments conducted on three benchmark demonstrate that S2DN surpasses performance state-of-the-art models. These effectiveness consistency enhancing robustness filtering contaminated