互联网文本蕴含道路交通信息抽取的模式匹配方法  

A Pattern Matching Method for Extracting Road Traffic Information from Internet Texts

作　　者：仇培元[1,2] 张恒才[1] 陆锋[1] 

机构地区：[1]中国科学院地理科学与资源研究所资源与环境信息系统国家重点实验室,北京100101 [2]中国科学院大学,北京100101

出　　处：《地球信息科学学报》2015年第4期416-422,共7页Journal of Geo-information Science

基　　金：国家"863"计划课题(2012AA12A211;2013AA120305)

摘　　要：互联网页面和社交网络文本中蕴含丰富的道路交通信息,是其他交通信息采集平台的有效补充。然而,自然语言文本形式的交通信息多以线性参考或地标方位描述交通事件空间位置,且大量存在事件元素缺失或隐含现象,对交通信息的自动化抽取有着较大影响。考虑到交通信息的自然语言表达方式虽然自由随意,但表达模式相对固定,提出一种从互联网文本中抽取道路交通信息的模式匹配方法。首先,基于道路交通事件描述的语言特征构建模式库;然后,以特征词词性序列的形式表达互联网文本和抽取模式,利用DTW距离度量序列相似度,实现抽取模式匹配;最后,在匹配抽取模式和填补规则指导下获取结构化的道路交通信息。由上海市城市交通相关门户网站和微博客平台的实验过程显示,本文所提出的模式匹配方法,抽取道路交通信息的准确率和召回率分别达到90%和80%以上,表明该方法能有效抽取互联网文本蕴含的道路交通信息,且实现过程相对简单,易于扩展,具有可用性。Internet pages and microblog messages usually contain a great amount of road traffic information that can become an important data source for city road traffic collection. However, current information extraction technology for Chinese natural language text is not applicable to extract road traffic information from Internet texts for two reasons： （1） the location descriptions in these texts are usually in the form of linear reference methods; and （2） some information elements are missing or ignored in the expressions. In this paper, we propose a pattern matching method for extracting road traffic information from Internet texts. This method focuses on obtaining the location element and event element of road traffic information, due to the fact that these elements are often associated with the above issues. Firstly, extraction pattern is defined as a sequence in which each item contains two parts： part of speech （POS） of the road traffic feature words, and information attribute type. Then an extraction pattern library is established based on the linguistic features of the road traffic event description. Secondly, the Internet text after pre-progressing and the extraction patterns are both represented by POS sequences. Thirdly, the method of measuring similarity between sequences with dynamic time warping （DTW） theory is used in pattern matching to look for the most suitable extraction pattern for this text from the library. Finally, the elements and attributes of traffic information are extracted from the text under the guidance of the matching pattern. To add the missing or ignored elements, special filling rules based on the syntactic structure of information expression are introduced into this extraction process. In an experiment that takes relevant Intemet texts for road traffic in Shanghai as the test data, whose sources are mainly from the official traffic information websites and Sina microblog platform, the precision and recall rate of road traffic information extraction is analyzed to b

关 键 词：互联网文本 道路交通信息 模式匹配 DTW距离 信息抽取 

分 类 号：P208[天文地球—地图制图学与地理信息工程]