Computational Behaviour Analysis (CBA) is an emerging interdisciplinary research area that draws equally from computer science and the study of human behaviour. It is intended to develop computational models and methods to represent and analyse human behaviour and their dynamics – a key element towards creating a digital twin – a digital replica of a human. Its ultimate purpose is to quantitatively assess the quality of human behaviour, identify long-term patterns and predict behaviour trajectories, thus recognising changes and potential behaviour projection. CBA plays a critical role in digital health by allowing for continuous, reproducible, and more objective assessments of human behaviour, and simulation-based condition prediction, thus enabling automatic detection of onset or progression of medical conditions, e.g. mental illness, the provision of personalised and adaptive assistive living, e.g. self-care/management.
CBA is built upon but goes beyond activity modelling and recognition. Such a quantitative assessment of the quality of relevant behaviours essentially corresponds to the analysis of how (well) activities are performed and has the quality of these activities changed. On top of CBA, digital twinning is a more holistic approach to creating a complete digital model of human in digital world which can imitate and manifest the exact same behaviours as humans in the real world. In this context the capabilities of existing approaches for modelling human activities, i.e., data mining and machine learning-based approaches, and domain and prior knowledge based approaches, are rather limited. For analysing human behaviour and their dynamics digital twinning requires: (i) robust bootstrapping techniques for model estimation that draw from both domain knowledge and task-specific sample data at different levels of abstraction; (ii) adaptation techniques for data-driven personalization of statistical behaviour models; (iii) behaviour dynamic modelling to capture and model the changing nature of behaviours; and (iv) approaches for unsupervised modelling of “normal” behaviour and automatic detection of deviations from it. Nevertheless, behaviour dynamic modelling has so far received little attention; the research is still in its infancy.
This project will bridge the aforementioned knowledge gap by developing (a) an enhanced behaviour model for a specific dimension of behaviour like physical activities, or social interactions; (b) the evolution mechanisms of the model to capture behaviour dynamics from specific angles, e.g. the different age cohorts or different severity of conditions; (c) model based behaviour simulation and trajectory prediction methods. Central to the above research is the development of core digital biomarkers which can best characterise human behaviour and their dynamic changes. Subsequently they will be used for change detection and projection of future behaviour and consequences for various application scenarios.
Built upon extensive research expertise and strengths on activity recognition and behaviour analysis, this project advances the research frontier by initiating a new investigation of behaviour dynamics towards digital twinning. It is in line with Ulster’s strategic research directions under ongoing initiatives such as data analytics, digital health and AI Centre of excellence. The project is expected to generate high-value scientific outputs in top-tier journals and provide inputs to research grant applications.