Analysis of Active Learning for Social Media to Support Crisis Management

Abstract

People use social media (SM) to describe and discuss different situations in which they are involved, such as crises. It is therefore useful to exploit SM content to support crisis management, especially by revealing useful and unknown information about real-time crises. Therefore, we propose a new active online multi-prototype classifier called AOMPC. It identifies relevant data related to a crisis. AOMPC is an online learning algorithm that runs on data streams and contains active learning algorithms for actively querying the label of obscure unnamed data. The number of queries is controlled by a consistent budget strategy. In general, AOMPC allows for somewhat labeled data streams. AOMPC evaluated using two types of data: (1) synthetic data and (2) SM data from Twitter related to two crises, the Colorado floods and the Australian pushfires. To provide a complete estimate, a complete set of known measurements was used to examine the quality of the results. Furthermore, a sensitivity analysis was conducted to show the effect of the parameters of AOMPC on the accuracy of the results. AOMPC's comparative study was conducted against other available online learning methods. Tests to handle emerging, somewhat labeled data streams showed AOMPC's excellent behavior.