Catching crime: detection of public safety incidents using social media

Abstract

The increasing prevalence of Social Media platform use has brought with it an explosion of new user generated public data. This data is centered around many, diverse topics. One theme of interest is how one can tap into the public safety and crime related user generated data to better understand patterns in the occurrence of crime incidents. One challenge in such data is that most of the data needs human annotation to make it usable by machines to analyse. This paper explores how different features, extracted from social media data, impact the performance of different classifiers. The classifiers are built to classify social media data as having to do with a reported crime or not. The challenge of few labelled data is discussed as well as different approaches to extracting features from the text data as well as the graph created by users interacting with each other is explored.