Bryan Maddox and Peter Andrews

Abstract

We describe a study that investigates how examiner behaviour in digital marking can be inferred from data on response processes, rather than relying on self-reported marking “strategies”. We capture marker behaviour between the clicks, with eye-tracking data from 25 participants completing 30 marking trials. We modelled attention and judgement behaviours with a Graph Neural Network and Transformer-based architecture. Our findings reveal micro-behaviours shaped by interface design and marking content. We draw conclusions on how improved Human–AI Interaction for assessment may reduce marker cognitive load, to support fairer, more reliable assessment. These findings contribute to a wider research initiative for Human-AI interaction under the EPSRC Prosperity programme, a collaboration between AQA and Kings College London.