Neubauer, Jessica Lilly; (2024) The Use of Computational Text Mining Methods to Detect and Understand Domestic Abuse. Masters thesis (M.Phil), UCL (University College London).

Preview

Text Neubauer_10188930_ Thesis_Redacted.pdf Download (2MB) | Preview

Abstract

Domestic Abuse (DA) is a widespread problem which causes major harm to victim-survivors. Psychological abuse is a common form of DA, and has a significant negative impact on victims, but is still not well defined or understood. Existing survey-based methodologies for researching psychological abuse could be complimented by computational social science methodologies using social media data. This thesis discusses the use of computational text mining in DA research, and seeks to contribute to this work through the creation of a dataset and a machine learning classifier to identify types of psychological abuse. A systematic literature review was conducted to give an overview of current work applying text mining methodologies in the study of DA and identified a gap in the literature regarding automatic identification of psychologically abusive behaviour. A dataset (n=2000) of Reddit posts was developed and labelled using an annotation scheme of six types of psychologically abusive behaviour. The annotation scheme was developed based on a literature review of existing measures and frameworks for psychological abuse, and refined over a series of expert discussions. Finally, a variety of machine learning classification models were trained on the dataset of psychologically abusive behaviours. A DistilBERT pre-trained model performed well (F1- score = 0.81) at classifying Threatening, Intimidating and Punishing behaviour. However, machine learning models were not successful at classifying other types of psychological abuse, due to the small size of the dataset and highly imbalanced classes. The thesis demonstrates that computational text analysis tools are useful for analysing large amounts of text data about DA, and providing insights into experiences of abuse that go beyond traditional qualitative methods. However, the thesis also illustrates the limitations of computational methods, which struggle to work well in the context of wider disagreements and debates about what constitutes abuse.

Download activity - last month

Export as JSONXMLCSV

Download activity - last 12 months

Export as XMLJSONCSV

Downloads by country - last 12 months

Export as XMLJSONCSV

Archive Staff Only

View Item