军事领域中概念抽取及其应用

Concept Extraction and Its Application in Military Domain

在军事领域的概念抽取中,以军事领域核心概念作为种子概念扩充了领域概念,并提出了基于核心概念结构的候选概念评价方法。该方法利用核心概念表征的不同语义范围,对语料进行层次聚类,计算不同层次细粒度的统计特征,对候选概念的领域归属度进行评价,避免语料分布不均匀造成候选概念的归属度差异。试验表明,该方法在一定程度上提升了准确率和召回率,通过不同细粒度的领域相关度和领域一致度统计特征,优化候选概念的抽取和排序,可应用于概念集扩充和概念推荐等场景。

Aimed at concept extraction in military domain, a domain concept is expanded by tak-ing the core concepts of military domain as the seed concepts. An evaluation method for candidate concepts based on structure of the core concept is proposed. Utilizing different semantic ranges represented by the core concepts, the method clusters the corpuses in the hierarchies and calcu-lates fine-grained statistical characteristics in the different hierarchies. Then, the domain consis-tency and the relevancy of the candidate concepts are evaluated to avoid the gap caused by uneven corpuses. Experimental results show that the method improves the precession and the recall rate. With the different fine-grained statistical characteristics of the domain consistency and the rele-vancy, the method can optimize the extraction and the sorting of the candidate concepts, thus it can be applied in the concept set expansion and the concept recommendation.