Hammocks, Daniel; (2025) Policing the Future: Horizon Scanning Through Computer Automated Information Prioritisation. Doctoral thesis (Ph.D), UCL (University College London).

Text Hammocks_10216321_thesis.pdf Access restricted to UCL open access staff until 1 November 2026. Download (15MB)

Abstract

What is crime? What is a “new” crime? and, what makes one crime “different” from another? The first part of this thesis explores crime from a unique innovative perspective assessing the societal, political and practical implications on how crime is defined whilst exploring the lack of consensus demonstrated in existing publications. Underlying definitions of the crime domain are stripped back and redefined with a futures-oriented perspective in mind whilst distinguished phenomena such as modus operandi and crime scripts are challenged and updated. This facilitates the proposal of a new methodology for recording crimes as well as a series of mechanisms of modus operandi evolution (MoMOE) based on a repurposed framework for differentiating between transition and emergence versus true novelty. All with the aim of answering; what is novelty in the crime domain? These conceptual foundations are then built upon from a technical perspective with the intention of crafting a complete Automated Horizon Scanning pipeline. Composed of three studies, this thesis aims to extract information pertaining to the commission of crime, or so-called modus operandi attributes, from unstructured free-text crime reports using a variety of natural language processing techniques such as fine-tuned reading comprehension models, semantic role labelling, and large language models. These features, alongside the full crime report, are then used to identify potentially new and emerging methods of perpetration through semantic anomaly detection and online hierarchical topic modelling. A key component of this research is the creation of a crowd-sourced dataset of crime reports, each annotated with their constituent modus operandi attributes, to overcome limitations on working with sensitive police data. The labelling accuracy was assessed through a secondary comparative annotation task and the dataset used to evaluate the models employed before trialling the final stage on a synthetic as well as real-world dataset provided by Greater Manchester Police.

Download activity - last month

Export as XMLJSONCSV

Download activity - last 12 months

Export as JSONXMLCSV

Downloads by country - last 12 months

Export as CSVXMLJSON

Archive Staff Only

View Item