一种基于句子结构特征的领域术语上下位关系获取方法

针对领域本体构建中概念上下位关系获取难的问题，提出融合句子结构特征的概念上下位语义关系抽取方法。首先利用层叠条件随机场（cascaded conditional random fields, CCRFs）算法建模实现概念上下位实体识别，然后通过对句子结构特征分析得出融合概念上下位关系的句子结构特征，最后利用融入句法特征基于支持向量机（support vector machine，SVM）建模的方法实现概念上下位关系抽取。为验证提出方法的有效性，以旅游领域上下位实体关系抽取为例进行了相关实验。实验结果表明：基于CCRFs模型的识别效果相对于现有的单层模型有较大改进, 其F值提高了6.57％；加入句法特征基于SVM概念上下位关系抽取方法较现有的基于条件随机场(conditional random fields，CRFs)概念上下位关系抽取方法更有效，其F值提高了4.68％。

An acquisition method of domain-specific terminological hyponym based on structure features of sentence

Aimed at the difficulty of obtaining the hyponymy relations of concept in the construction of domain ontology, a method based on the structure features of sentence for domain-specific terminological hyponymy extraction is proposed. Firstly, the CCRFs (cascaded conditional random fields) algorithm is used to recognize the entity of hyponymy. Secondly, by analyzing the sentence structure characteristic, the sentence structure feature fusing concept hyponymy is obtained. Finally, a method based on SVM (support vector machine) with the structure features of sentence is used to extract the hyponymy relations of domain-specific terminological. In order to verify the effectiveness of the proposed method, some tests are given to gain hyponymy entity relations of the tourism concept. Experiments results show that the value F of the classifier based on CCRFs is bigger than that of the classifier based on CRFs (Conditional Random Fields) by 6.57 percent and the value F of the hyponymy extraction method based on SVM with the structure features of sentence is bigger than that of the method based on CRFs(conditional random fields) by 4.68 percent. It is obvious that the proposed method in this paper is more effective than the existing methods.