Text mining police narratives to identify types of abuse and victim injuries in family and domestic violence events

Abstract

Police attend numerous family and domestic violence (FDV) related events each year and record details of these events as both structured data and unstructured free-text narratives. These descriptive narratives include information about the types of abuse (eg physical, emotional, financial) and the injuries sustained by victims. However, this information is not used in research. In this paper we demonstrate the application of an automated text mining method to identify abuse types and victim injuries in a large corpus of NSW Police Force FDV event narratives (492,393) recorded between January 2005 and December 2016. Specific types of abuse and victim injuries were identified in 71.3 percent and 35.9 percent of FDV event narratives respectively. The most commonly identified abuse types mentioned in the narratives were non-physical (55.4%). Our study supports the application of text mining for use in FDV research and monitoring.