计算机工程与应用 ›› 2023, Vol. 59 ›› Issue (8): 167-174.DOI: 10.3778/j.issn.1002-8331.2112-0193

• 模式识别与人工智能 • 上一篇    下一篇

融合词信息嵌入的注意力自适应命名实体识别

赵萍,窦全胜,唐焕玲,姜平,陈淑振   

  1. 1.喀什大学 计算机科学与技术学院,新疆 喀什 844008
    2.山东工商学院 信息与电子工程学院,山东 烟台 264005
    3.山东工商学院 计算机科学与技术学院,山东 烟台 264005
    4.山东省高等学校协同创新中心:未来智能计算机,山东 烟台 264005
  • 出版日期:2023-04-15 发布日期:2023-04-15

Attention Adaptive Model with Word Information Embeding for Named Entity Recognition

ZHAO Ping, DOU Quansheng, TANG Huanling, JIANG Ping, CHEN Shuzhen   

  1. 1.School of Computer Science and Technology, Kashi University, Kashi, Xinjiang 844008, China
    2.School of Information and Electronic Engineering, Shandong Technology and Business University, Yantai, Shandong 264005, China
    3.School of Computer Science and Technology, Shandong Technology and Business University, Yantai, Shandong 264005, China
    4.Co-innovation Center of Shandong Colleges and Universities:Future Intelligent Computing, Yantai, Shandong 264005, China
  • Online:2023-04-15 Published:2023-04-15

摘要: 缺少分词信息及未登录词、无关词干扰是字符级中文命名实体识别面临的主要问题,提出了融合词信息嵌入的注意力自适应中文命名实体识别模型,在新词发现的基础上,将字向量嵌入与词级信息嵌入融合作为模型输入,减少了未登录词对模型的影响,并增强了实体特征的显著性,使实体特征更容易被学习器获取;同时,在注意力机制中引入动态缩放因子,自适应地调整相关实体和无关词的注意力分布,一定程度上减小了无关词对模型的干扰。将该方法在公共数据集上进行实验,实验结果证明了方法的有效性。

关键词: 中文命名实体识别, 注意力机制, 动态缩放因子, 未登录词

Abstract: Lack of word segmentation information, interference of out of vocabulary words and irrelevant words are the main problems faced by character-level Chinese named entity recognition. In this paper, an attention adaptive model with word information embeding for Chinese named entity recognition is proposed. Based on the discovery of new words, the integration of character vector embedding and word-level information embedding is used as the input of the model, which reduces the influence of out of vocabulary words on the model, enhances the significance of entity features and makes them easier for learners to acquire. At the same time, the dynamic scaling factor is introduced into the attention mechanism to adaptively adjust the attention distribution of related entities and irrelevant words, which reduces the interference of irrelevant words to the model. The experimental comparison of this method on public datasets proves the effectiveness of the method.

Key words: Chinese named entity recognition, attention mechanism, dynamic scaling factor, out-of-vocabulary words