面向中医古籍的隐式关系抽取方法研究

Research on Traditional Chinese Medical Text Implicit Relation Extraction Method

自然语言种类丰富、形式灵活多变的特征使得隐式关系抽取成为目前关系抽取领域中富有难度和挑战性的任务之一。通过引入构式语法理论和依存句法分析两种认知语言学范畴的理论技术,构建了一种面向中医古籍中隐式关系的抽取方法。首先利用构式语法理论制定文本构式化策略、分析并定义出8种构式特征与5种构式类型,并使用CART(classification and regression tree,CART)分类模型完成文本分类;其次对其中4类构式使用依存句法分析技术构建句法树,通过分析句法树中的特定结构,制定医学类实体间的关系三元组抽取规则,实现隐式关系抽取;最后在经典中医古籍《黄帝内经》数据集上进行测试,实验结果表明了方法的有效性。

The variety of natural languages and their flexible forms are very rich,which makes implicit relation extraction one of the difficult and challenging tasks in the field of relation extraction.Two theoretical techniques were introduced in the field of cognitive linguistics,namely constructive grammar theory and dependent syntactic analysis,to construct a method for extracting implicit relations in traditional Chinese medical texts.Firstly,the constructive grammar theory was used to formulate a text structuring strategy,analyze and define eight constructive features and five constructive types,and the CART classification model was used to classify the text.Secondly,the dependent syntactic analysis technique was used to construct a syntactic tree for four constructs,and by analyzing the specific structure of the syntactic tree,the extraction rules of the relational triad between medical entities were formulated to realize the implicit relation extraction.Finally,the tests were conducted on the dataset of the classic traditional Chinese medical text,i.e.,Huangdi Neijing,and the experimental results showed the effectiveness of the method.