Potentially harmful, probably harmlessDefining and reducing false positives in digital risk assessment

Abstract: An area of research that has attracted more and more attention lately is digital risk assessment. This is due to its promising potential of identifying lone offenders, who often utilise digital platforms to express their thoughts and opinions. This thesis aims to improve the General Risk Score (GRS), a type of digital risk assessment used to identify lone offenders online, by reducing the number of false positives. A top-down approach in text analysis by way of dictionaries as well as a classification analysis (ROC) was applied to 71 texts drawn from a sample (N = 52 542) and then tested against another sample (N = 54 765). The results showed that whilst it was possible to significantly reduce the false positives, the true positives were also affected. Moreover, there were some interesting results that may inspire future research, such as the use of “I”-words (words referring to the self) and how they to a certain extent can indicate whether a lone offender or the general population wrote the text being analysed.