Predicting Targeted Violence from Social Media Communication

Abstract

For decades, threat assessment professionals have used structured professional judgment instruments to make decisions about, for example, the likelihood of violent behavior of an individual. However, with the increased use of social media, most people use online digital platforms to communicate, which is also the case for potential violent offenders. For example, many mass shootings in recent years have been preceded by communication in online forums. In this paper, we introduce methods to identify markers of the warning behaviors Leakage, Fixation, Identification, and Affiliation and examine their discriminant validity. Our results show that violent offenders score higher on these markers and that these markers were present among a significantly higher proportion of violent offenders as compared to the normal population. We argue that our method can be used to predict potential planned, purposeful, or instrumental targeted violence in written communication. Automated methods for detecting warning behavior from written communication can serve as a complement to traditional threat assessment and provides unique opportunities for threat assessment beyond traditional methods.