面向问答的数值信息抽取

Numerical Information Extraction for Question Answering

数值信息作为数据的一种直观表达方式,能够真实、有效地表达数据的量化含义.完整的数值信息要素包含有数字,数字的量化对象,如:主体及其属性,以及时空约束条件.对文本中蕴含的数值信息进行有效识别和抽取,并以问答的形式反馈给用户是大数据挖掘的重要意义,越来越受到学术界和产业界的关注.提出了一种基于数值模板结合条件随机场的数值信息抽取方法,该方法利用数值相关知识来制定数值信息抽取模板,对文本模式匹配,并使用条件随机场修补模板的局限性.实验结果的准确率、召回率和F值分别为0.847、0.665和0.745.

As an intuitive expression of data, numerical information could display the quantitative meaning of the data in a real and effective way. Quantized objects of numbers and numbers were included in the complete numerical information, such as the subject and its attributes, as well as space-time constraints and so on. Numerical information was effectively identified and extracted from the text. And fed back to the user in the form of question and answer was an important part of large data mining, which was paid more and more attention by academia and industry. Based on numerical template combined with conditional random fields, a numerical information extraction method was proposed. Knowledge related to numerical values was used in this method to develop numerical information extraction template for extracting numerical information. The conditional random fields was used to repair limitations of template. The accuracy, recall, and F value of the experimental results were 0.847, 0.665 and 0.745 respectively.