基于文本向量的古代针灸知识谱系构建研究

The ancient acupuncture and moxibustion knowledge resources are abundant, not only contained in ancient books of acupuncture and moxibustion, but also scattered in the mass of Chinese ancient books. But systematic collation and excavation has not been carried out in long-term. The perspective of knowledge pedigree is different from the general sense of the literature review, focusing on the formation process of knowledge system and the relationship between different knowledge, which is neglected in previous studies. From the complicated ancient acupuncture and moxibustion literature, a clear knowledge of genealogy could be combed out, which is conducive to deepening understanding the regular pattern of the formation and development of acupuncture knowledge, and a more comprehensive understanding of the scientific and historical connotation of acupuncture theory..The core of the pedigree construction of ancient acupuncture and moxibustion knowledge is to discriminate the time source and derivation relationship and spatial relationship from a large number of knowledge texts. However, traditional literature research and theoretical analysis are in vain in dealing with rich ancient acupuncture knowledge resources. Key factor of text space vectors should have been studied by using K-means, hierarchical clustering, and complex network technology in order to construct ancient acupuncture and moxibustion knowledge pedigrees that fully reflect the inheritance and evolution of acupuncture knowledge based on the completed research of ancient scattered acupuncture knowledge framework and scattered acupuncture literature, in which the origin of ancient acupuncture knowledge, and ancient acupuncture tree inheritance and multiple inheritance relations should be combed out. In addition, the consideration of theoretical analysis and academic history method will be also key factors in this research. The pedigree will fully reflect the inheritance and evolution of acupuncture knowledge and promote the comprehensive understanding and deep use of acupuncture knowledge.

古代针灸知识资源丰富，既载于针灸古籍，又散见于海量中医古籍，但长期缺乏系统整理与深入挖掘。知识谱系的视角不同于一般意义的文献梳理，着眼于知识体系形成过程及不同知识间的关联，在以往研究中较为忽略。从庞杂的古代针灸文献中，梳理出清晰的知识谱系，有利于深入把握针灸知识形成发展规律，更全面理解针灸理论的科学、历史内涵。.古代针灸知识谱系构建的核心在于从大量知识文本中辨析时间上的源头与派生关系、空间结构关系，但面对丰富的古代针灸知识资源，传统文献研究与理论分析难免力有不逮。本研究拟基于前期古代散在针灸知识框架研究及散在针灸文献整理，抓住文本空间向量计算这个关键因素，运用K-means、层次聚类、复杂网络技术，确定古代针灸知识谱系源头，梳理古代针灸知识谱系的树形继承、多重继承关系，在理论分析、学术史考量关照下，构建能充分反映针灸知识传承关系与衍变面貌，促进针灸知识全面学习与深度利用的古代针灸知识谱系。

古代针灸知识谱系是反映知识内容形成与衍变的重要途径，但长期以来多偏重于针灸理论内涵研究与针灸文献整理，缺乏对知识层面的系统梳理。本项目在前期古代散在针灸知识框架和散在针灸文献研究的基础上，运用信息技术方法结合传统理论文献研究，来解决对繁杂的古代针灸知识内容的系统考察，并将之还原为文本向量计算这个核心信息技术问题。通过构建文本向量计算SimCSE模型，计算相关文本内容的语义相似度，构建相似度矩阵，并对文本之间的关系进行筛选，保留较高相似度的文献，结合实体之间的关系，共同构成异构的复杂网络，继而通过社区发现算法，对文献进行聚类，从而获得文献传承关系，运用Tensorflow Embedding Projector对SimCSE生成的向量进行了可视化表示。.基于文本向量计算结果，按照四个要素的原则：知识源头、知识重要节点、知识异化点、知识主要分支，从三个层面构建古代针灸知识谱系，主要体现为：概念术语层面，以较为重要的601个古代针灸概念术语为主，以“古代针灸知识谱系源头知识集”为主要形式；重要表述层面，以短语或句子的形式对针灸知识内容的论述为主，包括132条内容，以“古代针灸知识谱系树形关系集”为主要形式；文献段落层面，以数个句子或段落的形式对针灸知识内容的论述为主，涉及53部重要古代针灸文献，以“古代针灸知识谱系多重继承关系集”为主要形式。.本项目所构建的古代针灸知识谱系，丰富了针灸理论文献研究领域，有利于古代针灸知识更清晰的表示，有助于其更好地实现当代理解、传承，促进其更广泛而深度的利用。