A collaborative approach to applying Natural Language Processing (NLP) to Domestic Homicide Reviews (DHRs): A study protocol.

Since 2011, there has been a statutory requirement in England and Wales to conduct a Domestic Homicide Review (DHR) into any domestic abuse-related death: a multi-agency review into the death of a person aged 16 or over that appears resulting from violence, abuse or neglect from an intimate partner, family member or household member. However, analyses of large numbers of DHRs are rare. One of the core challenges is the time and effort required to analyse narrative text within reports. Doing so manually is both time-consuming and resource-intensive and is a primary reason why researchers typically focus on only a portion of the available data. Natural Language Processing (NLP)-a sub-branch of artificial intelligence that enables computers to interpret and process natural language-provides a viable and scalable alternative by offsetting much of the heavy data processing to a computer. This protocol outlines a study to assess the feasibility of applying NLP to DHRs. The study will take a collaborative approach which balances the speed and scale of automation with the embedded knowledge and expertise of practitioners. This approach helps to ensure that outputs of NLP are sensitive and transparent about the biases common within datasets on violence and abuse. Based on initial consultations, we identify the priority research questions for investigation. In addition, we outline details of an ongoing collaboration with one partner, Imkaan. The protocol describes the data access, and retrieval and analysis stages before summarising how feasibility will be evaluated. We anticipate that several challenges will emerge within this study and identify strategies for mitigation. We suggest that working with practitioners who hold deep contextual knowledge about the social realities of violence and abuse, including language, risks, and experiences, mean that tools can be developed that are accountable to communities and appropriately applied to real-world problems.