面向数字人文的典籍语义词汇抽取研究——以SikuBERT预训练模型为例

Study on Keyword Extraction from Ancient Chinese Classics in the Context of Digital Humanities:Taking SikuBERT Pre-training Model for Example

利用关键词抽取技术可以帮助读者提取高度凝练的文本主题,快速获得古籍文本的中心内容,对普及古汉语知识和传播中华优秀传统文化有着重要意义。文章以先秦两汉时期的“儒家”“史书”数据库语料为分析对象,测试SikuBERT预训练模型在古汉语文本关键词抽取任务中的性能。实验结果表明,从两类语料中分别抽取的20个关键词基本能体现出相应典籍类别的主题内容:前者与所抽儒家典籍文本的相似度为76%~78%,后者与所抽史书文本的相似度为75%~78%。实验验证了SikuBERT模型在古籍文本关键词抽取中的适用性,能为开展古汉语文本主题分类、聚类和知识深度开发提供参考。

Keyword extraction technology could help readers to obtain the themes as well as the core content of highly-condensed ancient Chinese classics more effectively,which is of great significance to the popularization of ancient Chinese as well as the dissemination of traditional Chinese culture.With Pre-Qin and Han Confucian works and historical works as the corpus,this paper tries to testify SikuBER’s performance in keyword extraction from ancient Chinese classics.The result shows that the 20 keywords extracted from Pre-Qin and Han Confucian works bear a similarity of 76%~78%to the themes of the target texts,while the 20 keywords extracted from Pre-Qin and Han historical works bear a similarity of 75%~78%to the themes of the target texts.Since its applicability in keyword extraction from ancient Chinese classics is verified,SikuBERT could be further used in the thematic classification,text clustering and knowledge mining of ancient Chinese texts.