从中文Web网页中获取实体简称的研究

Extracting Abbreviated Names for Chinese Entities from the Web

简称是自然语言词汇的重要组成部分,其获取是自然语言处理中的一个基本而又关键的问题。提出了一种根据汉语全称从Web中获取对应汉语简称的方法。该方法包括获取和验证两个步骤。获取步骤通过选择查询模式从Web上获得候选简称集合。为了验证候选简称,定义了全简称关系约束,分别定性和定量地表示全称和对应简称之间的约束,构建了全简称关系图来表示所有全称和简称之间的联系,在验证过程中,先分别用约束公理和关系图对候选简称进行过滤,再用约束函数对候选简称分类,并以分类类别、语料标记和约束函数值作为属性构建决策树,利用决策树对候选简称进行验证。实验结果表明,获取方法的最终准确率为94.63%,召回率为84.09%,验证方法的准确率为94.81%。

Abbreviations are the essential parts of the vocabularies in natural language, therefore, acquiring abbreviations is a basic and significant task of natural language processing. We proposed a method of extracting abbreviations for the given Chinese full names from the Web. The method has two phases：candidate abbreviations extraction and verification. In the extraction phase,we constructed query items to issue to Google and saved the results as the corpora, from which we extracted candidate abbreviations. In the verification phase, we defined a full names/abbreviations relations constraints, which includes a group of constraint axioms and a group of constraint functions. We built a relation graph to reflect the connection of all the full names and the abbreviations. In the process of verifying, the incorrect ones could be filtered out using the constraint axioms and the relation graph the candidate abbreviations could be classified with the constraint functions, and the incorrect ones could be identified through a classifier, which was trained using the types of the candidate abbreviations, the values of the functions and the tags in the corpora. Comprehensive experiments show that the precision and recall rate of extracting abbreviation extraction are 94. 63% and 84. 09%, respectively, and the precision of candidate verification is 94. 81%.