Scholarlink

<p>Knowledge Graphs (KGs) are crucial in the field of artificial intelligence, supporting various applications such as search engines, intelligent recommendations, and sectors like architecture, finance, and healthcare. KGs are widely used for decision-making and information retrieval. Relation Extraction (RE) is an important technique for building KGs. However traditional supervised learning methods require a large amount of manually labeled data, resulting in lengthy processing times and high costs. Approaches based on remote supervision and semi-supervised learning face challenges with label noise. To overcome these issues, this study introduces a novel semi-supervised learning approach that models relationships between labels and instances using a unique label distribution for model training supervision. Furthermore, the contextual representation information is enriched by integrating entity location and label information. The proposed model outperforms strong baselines, achieving an average F1 score increase of 2.64&#xa0;% on the general-domain SemEval dataset and notably higher gains on the biomedical ChemProt dataset. These results indicate that the model remains robust under noisy supervision in different domains.</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

Enhanced semi-supervised relation extraction based on label confusion learning and multisource semantic aggregation

  • 期刊论文

  • 发表日期：

    2026年4月16日

查看全文
  • Lei Zhang,
  • Xinying Li,
  • Xiang Zhou

摘要

Knowledge Graphs (KGs) are crucial in the field of artificial intelligence, supporting various applications such as search engines, intelligent recommendations, and sectors like architecture, finance, and healthcare. KGs are widely used for decision-making and information retrieval. Relation Extraction (RE) is an important technique for building KGs. However traditional supervised learning methods require a large amount of manually labeled data, resulting in lengthy processing times and high costs. Approaches based on remote supervision and semi-supervised learning face challenges with label noise. To overcome these issues, this study introduces a novel semi-supervised learning approach that models relationships between labels and instances using a unique label distribution for model training supervision. Furthermore, the contextual representation information is enriched by integrating entity location and label information. The proposed model outperforms strong baselines, achieving an average F1 score increase of 2.64 % on the general-domain SemEval dataset and notably higher gains on the biomedical ChemProt dataset. These results indicate that the model remains robust under noisy supervision in different domains.