Using twitter for situational awareness after an earthquake: The roles of text categorization and location information

Twitter is an important platform to disseminate critical information for emergencies. Text categorization became a significant concept due to the rapid increase in the number of documents along with Twitter usage. In addition, it is important to know the location of the victims to respond to their needs, such as rescue operations, psychological assistance, and so on. The primary contribution of this thesis is that, it is a novel approach for text categorization and location information harvesting from earthquake-related tweets. The ultimate goal is to conduct sophisticated earthquake-related tweet analyses to achieve efficient and effective disaster management. Firstly, importance of location data in earthquake management is demonstrated by using location-specific tweets. Secondly, an uncapacitated p-median problem (UPMP) on Twitter data-set is applied to demonstrate important topics and related clustered tweets. To solve the UPMP problem, a novel hybrid genetic bat algorithm (HGBA) is introduced by using a similarity (distance) matrix gathered by the latent dirichlet allocation (LDA). Thirdly, to extract named entities such as location from tweets, named entity recognition tools in conjunction with recurrent neural network (RNN)-based approaches are used to ascribe a meaning to tweets in earthquake situations. Finally, Kernel density estimation (KDE) is used to identify hotspots based on important topics in each location. The proposed novel approaches are applied to analyze the Twitter data-set of the Nepal earthquake. All in all, this thesis focuses on extracting useful information from Twitter in case of an earthquake due to provide comprehensive earthquake management.