Computational models of hedonic evaluation: An explanatory-predictive framework

Hedonic evaluation underlies preferences across domains ranging from food and social stimuli to art. Increasing evidence suggests that these diverse judgements rely on common computational and neurobiological principles, motivating a unified perspective on how hedonic value is constructed. Computational modelling has become central to this endeavour by transforming conceptual theories into quantitative, testable accounts of how sensory information, prior experience, internal states and contextual factors give rise to subjective value. However, the rapid expansion of computational approaches has produced a heterogeneous landscape of models that differ in their assumptions, computational principles, explanatory ambitions and predictive performance. Here, we propose an explanatory-predictive framework for organising, comparing and evaluating computational models of hedonic evaluation. The framework characterises models according to two complementary dimensions: mechanistic explainability and empirical predictive performance. We also organise models into six broad families defined by their dominant computational principles. Finally, we argue that progress will depend less on competing modelling traditions than on integrating complementary computational principles into a general computational theory of hedonic evaluation.