My research is focused on building inductive models from data in settings where the interests of modellers and those being modelled are not aligned. This includes counterterrorism, law enforcement and policing, anti-money-laundering, fraud, and cybersecurity; but also areas that are less obviously adversarial, such as customer relationship modelling.
Two approaches that are especially useful because it is hard for adversaries to disguise their activities and traces are natural language and social networks. I have developed techniques to infer properties such as deception, intentions, personality, emotional state and attidues from language; and also ways to detect the processes that drive abusive language and hate speech, and the use of language for influence. I have also developed techniques to understand the structure of social networks, especially when the edges have different meanings (friends vs colleagues, friends vs enemies) and change with time.
My recent work includes inferring systemic nets from corpora, removing a major bottleneck to their wider use, and empirical determination of verbal mimicry.
My latest project is based on Measuring Human Emotion in Short Documents to Improve Social Robot and Agent Interactions.