基于多分类的地震信息跟踪方法  

Earthquake Information Tracking Based on the Multi-category Classification Method

作　　者：康建红[1] 庞晶源[1] 宁宝坤[2] 张晨侠[1] 

机构地区：[1]吉林省地震局,长春130117 [2]中国地震应急搜救中心,北京100049

出　　处：《华北地震科学》2015年第2期20-24,共5页North China Earthquake Sciences

基　　金：地震科技星火计划项目(XH14017Y);震情跟踪定向工作任务(2015010125)

摘　　要：信息跟踪是指在文本数据流中检测与某个突发事件相关的信息,突发事件的动态演化特性是造成信息跟踪漏报率高的原因之一。为此,提出了基于多分类思想的突发事件信息跟踪方法,即将突发事件演化过程中出现的多个焦点看成是多个分类,而多个分类会在跟踪过程中动态调整。在该跟踪方法中,提出了基于上下文和潜在语义分析的帖子表示模型以及基于多焦点的突然事件表示模型。最后将本文提出的方法应用到某特定地震的信息跟踪系统中。实验结果表明,潜在语义分析针对性地解决了论坛草根性带来的用词多样性问题,而多焦点的话题模型在一定程度上弥补了话题的动态演化特性对跟踪系统性能的影响,基于多分类思想的突发事件信息跟踪方法使得系统的漏报率和误报率有了不同程度的降低,改善了地震信息跟踪系统的性能。Information tracking is to detect the information related to a certain emergency from the text stream.The dynamic evolvement characteristic is one of the important reasons which lead to the high miss rate,and we propose an emergency information tracking method based on the multi-classification to deal with this problem,which regard the multi focuses appearing during the event evolvement as the multi classes.In this tracking method,we propose and explore a post representation model based on the context and latent semantic analysis and an event representation model based on the multi-focus.We apply the proposed information tracking method in the earthquake information,and the experimental results show that the proposed method can improve the performance of the earthquake information tracking system.

关 键 词：突发事件 话题跟踪 多分类 动态演化 

分 类 号：P315-391[天文地球—地震学]