结合事件和语境特征的台风事件信息抽取方法  

Typhoon Event Information Extraction Method Based on Event and Context Characteristics

作　　者：黄宗财 仇培元[2] 王海波 吴升[1,3] HUANG Zongcai;QIU Peiyuan;WANG Haibo;WU Sheng(Spatial Information Research Center of Fujian Province, Fuzhou University, Fuzhou 350003, China;State Key Laboratory of Resources and Environmental Information System, IGSNRR, CAS, Beijing 100101, China;Fujian Collaborative Innovation Center for Big Data Applications in Governments, Fuzhou 350002, China;Economic and Management School, Hubei University of Technology, Wuhan 430068, China)

机构地区：[1]福州大学福建省空间信息工程研究中心,福建福州350003 [2]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 [3]海西政务大数据应用协同创新中心,福建福州350002 [4]湖北工业大学经济与管理学院,湖北武汉430068

出　　处：《测绘科学技术学报》2019年第2期209-214,共6页Journal of Geomatics Science and Technology

基　　金：国家自然科学基金重点项目(41631177);数字福建建设项目(闽发改网数字函[2014]191号;[2016]23号;[2016]77号);福建省科技创新平台项目(2015H2001)

摘　　要：台风事件信息抽取在台风知识检索、灾害应急预警、损失评估预测等方面具有重要价值。由于台风事件存在一个时空动态演变的过程,新闻文本中易出现包括台风形成、登陆、移动和消散等不同时间节点的事件状态的混合表达,使得规则匹配和现有传统机器学习方法进行信息抽取效果不佳。因此,提出一种结合事件和语境特征的台风事件抽取方法,能够充分考虑词汇的事件特征和上下文信息。实验结果显示所提出的方法使得登陆时间和登陆位置抽取准确率高达92%,各事件元素抽取精度总体较高且较为平衡,平均F1值高达78%,提升幅度达42%。本文研究为构建台风知识库提供了方法支撑。Typhoon event information extraction has important value for typhoon knowledge retrieval, disaster emergency warning, loss assessment and prediction. Due to the dynamic evolution of typhoon events in time and space, there are often mixed expressions of event states of different time nodes in news texts, including typhoon formation, landing, movement and dissipation, which make rule matching and traditional machine learning methods ineffective in information extraction. Therefore, a typhoon event extraction method is proposed in the paper which combines event and context features. It can take full account of the event features and context information of vocabulary. The experimental results show that the proposed method makes the extraction accuracy of landing time and landing location as high as 92%, and the extraction accuracy of each event element is relatively high and balanced, with an average F1 value of 78% and an increase of 42%. This study provides a methodological support for the construction of typhoon knowledge base.

关 键 词：台风事件 新闻文本 事件语义特征向量 双向长短期记忆网络 条件随机场 

分 类 号：P208[天文地球—地图制图学与地理信息工程]