Interactive Learning for Identifying Relevant Tweets to Support Real-time Situational Awareness
Primary Investigator:
David Ebert
Luke S. Snyder, Yi-Shan Lin, Morteza Karimzadeh, Dan Goldwasser, and David S. Ebert
Abstract
First responders have increasingly used real-time social media data over recent years because of its ability to facilitate situational awareness and provide expediency in identifying crisis situations. However, due to the noise and deluge of data, effectively determining semantically relevant information can be difficult, further complicated by changes in topics over time. Although efforts have been devoted to improving short text relevance classification for crisis management, the majority of existing methods fail to incorporate user feedback into the classification process. As such, classifiers cannot be constructed and interactively trained for specific events or user-dependent needs. Existing methods that include interactive user feedback only do so for historical data as opposed to in real-time. This limits real-time situational awareness as streamed data that is incorrectly classified cannot be corrected immediately, permitting the possibility for important future data to be incorrectly classified as well. We present a novel interactive framework to improve the classification process in which the user iteratively identifies the relevancy of tweets in real-time to train the classification model on-the-fly for immediate predictive improvements. We provide and computationally evaluate a classification model adapted to learn at interactive rates. In addition, we implement our framework in the extended SMART 2.0 system, allowing for the user to interactively explore, filter, and refine the relevance of real-time Twitter and Instagram data.