Scholarlink

<p>Knowledge graph embedding (KGE) aims to embed entities and relations in knowledge graphs (KGs) into a continuous, low-dimensional vector space. However, traditional transductive methods require retraining when embedding out-of-knowledge-graph (OOKG) entities, which is inefficient. To address this issue, current research has shifted to inductive methods, which aim to learn a general embedding generation framework to infer OOKG entity embeddings. However, existing inductive methods only utilize local neighborhood information of OOKG entity, neglecting the global semantic association between entities. Type information, as shared semantic information among same-type entities, can be used to establish the global semantic association among the entities. Based on this, we propose an inductive KG embedding method called OOKG Entity Type Inference and Fusion Based on Neighborhood Structure(TIFNS). Specifically, TIFNS first classifies known entities by clustering based on the neighboring relations’ characteristics of known entities, further obtaining the type information of the known entities, and then leverages auxiliary triples and known entities’ type information to infer the type information of OOKG entity, finally fuses entity embedding with its type embedding to construct entity embedding with global semantic consistency. Notably, TIFNS is a pluggable method that can be easily integrated into existing models. Experimental results show that baselines-TIFNS achieve significant performance improvements in link prediction and triple classification tasks.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

An inductive KG embedding method for OOKG entity type inference and fusion based on neighborhood structure

  • 期刊论文

  • 发表日期：

    2026年1月29日

查看全文
  • Zhiming Li,
  • Guangkang Zhang,
  • Fude Liu,
  • Shuquan Zhou,
  • Heping Wei,
  • Dianlong You

摘要

Knowledge graph embedding (KGE) aims to embed entities and relations in knowledge graphs (KGs) into a continuous, low-dimensional vector space. However, traditional transductive methods require retraining when embedding out-of-knowledge-graph (OOKG) entities, which is inefficient. To address this issue, current research has shifted to inductive methods, which aim to learn a general embedding generation framework to infer OOKG entity embeddings. However, existing inductive methods only utilize local neighborhood information of OOKG entity, neglecting the global semantic association between entities. Type information, as shared semantic information among same-type entities, can be used to establish the global semantic association among the entities. Based on this, we propose an inductive KG embedding method called OOKG Entity Type Inference and Fusion Based on Neighborhood Structure(TIFNS). Specifically, TIFNS first classifies known entities by clustering based on the neighboring relations’ characteristics of known entities, further obtaining the type information of the known entities, and then leverages auxiliary triples and known entities’ type information to infer the type information of OOKG entity, finally fuses entity embedding with its type embedding to construct entity embedding with global semantic consistency. Notably, TIFNS is a pluggable method that can be easily integrated into existing models. Experimental results show that baselines-TIFNS achieve significant performance improvements in link prediction and triple classification tasks.