Automated categorization and mining tweets for disaster management

Abstract

Recent trend has shown that Twitter is an emerging source to monitor disaster events. This article aims at mining tweet messages to facilitate disaster manager for better response in times of emergency. The two main objectives here, are classification and extraction of tweets. In this work machine learning is applied on tweets that are generated during the course of a disaster, to categorize them into different stages of that disaster and extract the metadata. An initial effort has spent for manually labeling 650 tweet messages into predefined categories according to disaster phases to build the classification model. Thereafter, the trained model is used to automatically categorize the new subsequent tweets in a supervised way of machine learning. The five classes or categories are considered in this work are as preparedness, response, impact, recovery and other. Once the classification is done, next objective is to mine informations from the categorized tweet messages such as impacted locations, hardest hit areas, volunteer relief organization who are working on, whether that message is a complain, if any disease has spread etc. The result generated after classification and extraction, helps emergency manager so that they can take action quickly and efficiently in the impacted location.