Random Forest Location Prediction from Social Networks during Disaster Events

Rapid location and classification of data posted on social networks during time-critical situations such as natural disasters, crowd movement and terrorism is very useful way to gain situational awareness and to plan response efforts. Twitter as successful real time micro-blogging social media, is increasingly used to improve resilience during extreme weather events/emergency management situations, including earthquake. It being used during crises by communicating potential risks and their impacts by informing agencies and officials. The geographical location information of such events are vital to rescue people in danger, or need assistance. However, only few messages contains there native geographical coordinates (GPS). So identifying location is a real challenge with Twitter data during critical situations. Identification of Tweets and their precise location are still inaccurate. In this work, we propose to use semi-supervised technique to utilize unlabeled data, which is often abundant at the onset of a crisis event, along with fewer labeled data. Specifically, we adopt an iterative Random Forest fitting-prediction framework to learn the semi-supervised model.