采用术语定义模式和多特征的新术语及定义识别方法

Applying Terminology Definition Pattern and Multiple Features to Identify Technical New Term and Its Definition

新术语及其定义抽取是信息抽取的重要研究内容之一.研究结果表明,在科技文献中,一个新术语往往伴随其定义出现,通过考察,在真实文本中,术语定义存在显著的语言表述特征,从大规模真实语料库中,通过考察术语定义构成的语言学模式、定义中词汇和术语周边的统计特征,提出了以术语定义的语言学模式(LPTD)作为待识别候选新术语集,同时考虑到有关新术语出现的上下文统计特征,用SVM分类器方法完成科技语料中新术语及其定义的识别.在大规模科技期刊上进行方法验证,开放性评测结果的精确率为90.5%、召回率达78.1%.

identification of technical new term and its definition is an important research topic information extraction. It is still a great challenge to provide a scalable solution for large-scale terms extraction, because most previous approaches fail to explicitly define the linguistic constituent of terms and the function of their definition patterns. The authors＇ research shows that the occurrences of technical new terms in most cases are accompanied with their definition descriptions in the real corpus. Based on this intuition, the linguistic constituent of technical terms and the numerical function of their definitions are defined explicitly. Also presented is a novel statistical approach based on linguistic pattern of terminology definition （LPTD） to extract Chinese lechnical new terms and their definitions. LPTD in this paper is first proposed to delimit the boundary of technical terms. In the identification phase, both statistical information of terms and LPTD features obtained from the previous filtering process are taken into account in the SVM classifier. They are integrated into one unified framework. The idea in this paper can also be used for reference in collocation extraction （CE） and be easily extended to other different languages. Compared with the previously reported outcomes, this approach achieves a competitive result in real large-scale corpora at 90.5 % in precision and 78.1% in recall.