Twitter是一個快速成長的社群媒體之一,在Twitter中的每一則訊息被稱為「tweet」,限於140個字元,因為眾多的使用者來分享他們的觀察,tweets常常成為了報導真實世界中事件的即時訊息,特別是災害發生時,使用者常利用Twitter來報導何處需要幫助,這樣的tweets成為另一個潛在的資訊來源,並可支援災害救援。為了使用這些災害訊息容易被查詢、利用、和散佈,本文報告由tweet擷取災害救援資訊,並定位於線上地圖的方法和經驗。本文是以日本311地震後一個月內的tweets為研究對象,收集tweets之後,即進行具名實體辨識(Named Entity Recognition)將tweets中所提及的地名辨識出來,接著利用語彙的共現(co-occurrence)建立災害救援關鍵字的觀念階層(concept hierarchy),基於觀念階層可以整理出三種三元組結構以擷取災害救援資訊,災害救援資訊擷取後存入空間資料庫中,可便於存取且顯示於地圖,如Google Map,後端使用者(end users)因此可以在地圖上瀏覽災害救援資訊。
Twitter is a rapidly growing social media in recent years. The brief text sent via Twitter is called as ”tweets”, which is less than 140 characters. Since massive people post tweets to report their observations, tweets are often real-time sources reflecting a various kind of real-world events. In particular, people often post what/where they need when a disaster happens. These disaster tweets are potential information to support disaster relief. This study reports our experiences in extracting disaster information from tweets via natural language process. While collecting tweets, we use NER (Named Entity Recognition) technique to extract names of place and organization, and then apply co-occurrence analysis to develop concept hierarchies of disaster keywords. Based on these concept hierarchies, we build three types of triple-like structures for extract disaster relief information. Since the disaster relief information is stored in spatial database, it can be retrieved easily, and displayed on the map such as Google Map. End users, thus, can browse the disaster relief information on the map.