Scholarlink

<p>In the era of big data, knowledge graphs (KGs) have emerged as a powerful form of knowledge representation and are widely employed in applications such as question answering, recommender systems, and information retrieval. Despite their effectiveness, the construction of KGs often relies on automated extraction and manual curation, which can result in incomplete or missing information. This inherent limitation has motivated growing interest in knowledge graph reasoning, which aims to infer missing facts and enhance the utility of KGs in downstream tasks.The present paper proposes a novel framework for link prediction, which consists of a logic-semantic miner, a rule generation optimizer, and a logic reasoner. Specifically, the logic-semantic miner leverages large language models to capture potential semantic associations between predicates and performs logical validation to generate candidate rules; the rule generation optimizer enhances the stability and reliability of rule generation through multi-round generation and quality evaluation; finally, the logic reasoner conducts inference based on high-quality rules to achieve knowledge completion and link prediction. Experimental results demonstrate that the method proposed in this paper consistently outperforms baseline models across multiple datasets, significantly improving the performance of knowledge graph reasoning and providing new insights for the optimization and application of KGs. All the code and data in this work will be released at <a href="https://anonymous.4open.science/r/CpLoM-75D4.">https://anonymous.4open.science/r/CpLoM-75D4.</a></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

LRKG: integrating large language models with logical rule for knowledge graph reasoning

  • 期刊论文

  • 发表日期：

    2026年5月18日

查看全文
  • Kaiyue Ji,
  • Zili Zhou,
  • Peng Gao,
  • Zhaoai Wang

摘要

In the era of big data, knowledge graphs (KGs) have emerged as a powerful form of knowledge representation and are widely employed in applications such as question answering, recommender systems, and information retrieval. Despite their effectiveness, the construction of KGs often relies on automated extraction and manual curation, which can result in incomplete or missing information. This inherent limitation has motivated growing interest in knowledge graph reasoning, which aims to infer missing facts and enhance the utility of KGs in downstream tasks.The present paper proposes a novel framework for link prediction, which consists of a logic-semantic miner, a rule generation optimizer, and a logic reasoner. Specifically, the logic-semantic miner leverages large language models to capture potential semantic associations between predicates and performs logical validation to generate candidate rules; the rule generation optimizer enhances the stability and reliability of rule generation through multi-round generation and quality evaluation; finally, the logic reasoner conducts inference based on high-quality rules to achieve knowledge completion and link prediction. Experimental results demonstrate that the method proposed in this paper consistently outperforms baseline models across multiple datasets, significantly improving the performance of knowledge graph reasoning and providing new insights for the optimization and application of KGs. All the code and data in this work will be released at https://anonymous.4open.science/r/CpLoM-75D4.