About the project
Inter-disciplinary project to develop novel Large Language Model (LLM) based AI methods for classification of Digital Human Trafficking (DHT) behaviour in social media posts. It will explore human/AI teams for information foraging tasks to deliver a step change in capability for early detection and intervention of DHT into terrorism.
The UK is currently experiencing the highest levels of juvenile and youth terrorism-related activity in history. Facilitated by social media platforms, online recruitment of children and (particularly young) adults often relies upon deception, coercion, or false incentivisation to act in support of an extremist cause. Social media AI technologies in the form of Machine Learning and cognate models can enable forms of use that produce unintended harms.
This inter-disciplinary project will develop novel Large Language Model (LLM) based AI methods for Digital Human Trafficking (DHT) behaviour classification from social media posts, exploring if bringing together AI tools and human criminology experts in human/AI teams for information foraging tasks can deliver a step change in our capability for early detection and intervention of DHT into terrorism. A typical information foraging session might start with a wide-area focus (e.g. subject level extremist viewpoints) and iteratively narrow it down (e.g. a specific individual of interest) until the task is achieved.
The supervisor team includes both Computer Science (Prof Middleton with LLM expertise) and Criminology (Dr Vale with DHT into terrorism expertise) with secured access to VOX-Pol Network of Excellence and its large datasets of social media content suitable for training LLMs, including propaganda circulation and forum posts related to DHT into terrorism.
Both supervisors have good connections with UK law enforcement (e.g. National Crime Agency) and active grants (e.g. Middleton's ProTechThem grant ES/R003254/1 exploring few-shot NLP for online harms).
