返回Google圖書搜尋

Tracking and Extracting Public Response to Crisis on Twitter

出版	McGill University Libraries, 2015
URL	http://books.google.com.hk/books?id=OdILjwEACAAJ&hl=&source=gbs_api

註釋"Large-scale disasters can often impede conventional means of aggregating situa- tional awareness (SA), and therefore, obstruct efficient emergency response. Twitter is widely regarded as an alternate source of valuable SA, extraction of which has received significant interest. However, little attention has been given to the amount of useful information Twitter can provide and researchers continue to rely on hashtag- based keyword approaches for data collection. Our analysis reveals that disasters yield time-varying data streams with dissimilar SA, along with a noticeable absence of disaster-specific hashtags in tweets with novel information. We also find that these data streams take variable shapes that can be explained in terms of disaster characteristics: foreknowledge, duration, severity, and news media engagement. In order to tackle the challenges presented by the data, we introduce an adaptive filtering framework, tailored to the idiosyncrasies of an active Twitter stream. This frame- work can function as an integral component of any information extraction machinery that deals with live data. Our system utilizes a data model that implements a novel three-label classification scheme to describe the composition of the data stream. We evaluate our filters performance on simulated Twitter streams generated from this data model. The method is able to remove over 85% of non-relevant content, while achieving a three-fold increase in the recall of disaster-related tweets, compared to existing approaches. In combination, the method and the model are useful tools for extracting SA and highlight important directions for future work in this area."--