Antisocial Behavior corpus for harmful language detection

We report on experiments that demonstrate the relevance of our AntiSocial Behavior (ASB) corpus as a machine learning resource to detect antisocial behavior from text. We first describe the corpus and then, by using the corpus for training machine learning algorithms, we build a set of binary classifiers. Experimental evaluations revealed that classifiers built based on the ASB corpus produce reliable classification results with up to 98% accuracy. We believe that the dataset will be valuable to researchers and practitioners working in preventing, controlling and diagnosing antisocial behavior and related problems.