Scholarlink

<p>Medical knowledge graphs integrate heterogeneous medical knowledge and support information sharing and reasoning. However, conventional rule-based graph reasoning techniques typically rely on path induction or rule-feature learning, and do not adequately consider entity content variability and the influence of contextual subgraphs, which may lead to inappropriate rule transfer and unreliable inference. This paper proposes MedKGRR, a medical knowledge graph reasoning framework that incorporates entity–relation semantic embeddings into rule mining, to improve reasoning accuracy. The MedKGRR model combines the advantages of attention-based context modeling and logical rule learning by embedding entity-relationship semantic information into the rule mining process. Specifically, entity and relation encoders are used to obtain semantic representations, enabling context-aware rule decoding for reasoning. Experimental results on BBK and MED, two subsets of the public MED-BBK-9K dataset, show that MedKGRR achieves superior performance. On the more complex MED subset, the mean reciprocal rank (MRR) is 0.3742 and the Hit@10 is 58.02%, both of which outperform the baselines. Overall, these results indicate that the MedKGRR model is robust and adaptable for reasoning over medical knowledge graphs.</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

Research on entity relationship semantic embedded rule mining model for medical graph reasoning

  • 期刊论文

  • 发表日期：

    2026年5月20日

查看全文
  • Shuangbao Zhang,
  • Quan Cheng

摘要

Medical knowledge graphs integrate heterogeneous medical knowledge and support information sharing and reasoning. However, conventional rule-based graph reasoning techniques typically rely on path induction or rule-feature learning, and do not adequately consider entity content variability and the influence of contextual subgraphs, which may lead to inappropriate rule transfer and unreliable inference. This paper proposes MedKGRR, a medical knowledge graph reasoning framework that incorporates entity–relation semantic embeddings into rule mining, to improve reasoning accuracy. The MedKGRR model combines the advantages of attention-based context modeling and logical rule learning by embedding entity-relationship semantic information into the rule mining process. Specifically, entity and relation encoders are used to obtain semantic representations, enabling context-aware rule decoding for reasoning. Experimental results on BBK and MED, two subsets of the public MED-BBK-9K dataset, show that MedKGRR achieves superior performance. On the more complex MED subset, the mean reciprocal rank (MRR) is 0.3742 and the Hit@10 is 58.02%, both of which outperform the baselines. Overall, these results indicate that the MedKGRR model is robust and adaptable for reasoning over medical knowledge graphs.