Semantic tagging of heterogeneous data: Labeling Fire&Rescue incidents with threats

In the article we present a comparison of the classification algorithms focused on labeling Fire&Rescue incidents with threats appearing at the emergency scene. Each of the incidents is reported in a database and characterized by a set of quantitative attributes and by natural language descriptions of the cause, the scene and the course of actions undergone by firefighters. The training set for our experiments was manually labeled by the Fire Service commanders after deeper analysis of the emergency description. We also introduce a modified version of Explicit Semantic Analysis method and demonstrate how it can be employed for automatic labeling of the incident reports. The task we are trying to accomplish belongs to the multi-label classification problems. Its practical purpose is to support the commanders at a emergency scene and improve the analytics on the data collected by Polish State Fire Service.