Towards natural behaviour processing for the representation and prediction of digital behaviour

Characterising individual differences in behaviour and predicting future behaviour are central goals of psychological science. However, behavioural data are difficult to obtain, and most research relies on nomothetic approaches that aggregate them into summary statistics. This aggregation obscures person-specific patterns that would afford an idiographic understanding of individual behaviour. Pervasive digital technologies (e.g., smartphones, wearables, and computer-mediated systems) now generate fine-grained, temporally ordered records of everyday behaviour, creating unprecedented opportunities to study both systematic inter-individual differences and patterns of intra-individual variation in digital behaviour. Here, we introduce Natural Behaviour Processing (NBP) as a framework for analysing digital behavioural data using sequence modelling approaches adapted from natural language processing. We argue that behaviour shares structural properties with language: both involve a repository of possible actions (vocabulary), systematic patterns governing how elements typically combine (grammar), and context-dependent meaning (semantics). These structural parallels motivate applying sequence-based models, which have revolutionised language processing, to the study of behaviour more broadly. We provide a brief illustrative review of pioneering studies demonstrating that sequential approaches capture person-specific patterns that aggregation loses, across topics including life outcomes, decision making, social media use, consumer behaviour, digital education, and health. We discuss the challenges facing NBP adoption, including the tension between predictive modelling and theoretical understanding, the need for large-scale curated behaviour datasets, and methodological training requirements. By modelling the order and context of everyday actions rather than aggregates, NBP could enable person-specific representations of behaviour that support both individual-level prediction and systematic comparison of individual differences across people.