Skip to main content Skip to navigation

Centre for Operational Police Research

AI Systems for Understanding and Collecting Evidence in VAWG Investigations

Dr. Gabriele Pergola, in collaboration with Prof. Arshad Jhumka from the University of Leeds and a team of researchers from the University of Warwick, has partnered with the Forensic Capability Network (FCN) to develop an advanced NLP-based platform that enhances the investigation of VAWG cases. Funded by the STAR Fund, this project introduces a state-of-the-art tool designed to automate the detection and analysis of abusive language in digital communications, significantly boosting efficiency and standardisation in VAWG case reviews.

The research team engaged closely with law enforcement agencies and FCN personnel throughout the project, conducting pilot tests and gathering feedback from officers who specialise in VAWG cases. The platform enables rapid categorisation of messages, delivers contextual summaries, and offers persona profiling based on conversational data. A synthetic dataset was also developed to train the NLP models, ensuring that testing could proceed in a realistic setting without compromising sensitive case data.

Key outcomes of this project include:

  • Interactive Analysis Platform:Integrating NLP models into an accessible interface, the platform enables officers to upload, categorise, and visualise large volumes of conversation data. An AI-powered chatbot further supports interactive analysis and contextual understanding.
  • Conditioned Summarization:Specialized summarisation techniques allow the tool to focus on violence-related content, providing investigators with targeted insights for faster case assessment.
  • Synthetic Data Generation:Using large language models, the team created realistic, anonymised datasets that mirror real-world VAWG communication scenarios, allowing robust model training without privacy concerns.
  • Persona Profiling:The platform provides behavioural insights into individuals involved in VAWG cases, assisting officers in understanding interaction patterns and motivations.

The findings and developments from this project promote the standardisation of VAWG case handling across police forces, reducing reliance on manual review and expediting investigations. Future enhancements will focus on expanding data types to include image analysis, refining persona profiling capabilities, and integrating a dynamic feedback system for continuous improvement.

On the News: AI that detects hate messages against women could also be turned on drug dealers | The IndependentLink opens in a new window | Daily MailLink opens in a new window | MSNLink opens in a new window | FCN (2024)Link opens in a new window | FCN (2023)Link opens in a new window

Relevant Publications:

  • S. Khan, G. Pergola, and A. Jhumka, “Multilingual Sexism Identification via Fusion of Large Language Models”Working Notes ofCLEF2024.Winner of the competition for the English language.
  • L. Zhu, G. Pergola, L. Gui, D. Zhou, and Y. He, “Topic-Driven and Knowledge-Aware Transformer for Dialogue Emotion Detection,”Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics (ACL 2021),pp. 1571–1582, 2021.