Assessing the rereading effect of digital reading through eye movements using artificial neural networks.

Objective: This study aimed to investigate the differences in eye movement characteristics between first reading and rereading and to develop a neural network model for classifying these reading practices. The primary goal was to enhance the understanding of rereading identification and provide insights into assessing students' text familiarity.

Methods: We compared eye movement metrics during first reading and rereading, focusing on parameters such as total reading time, fixation duration, regression size, regression count, and local eye movement behaviors within areas of interest (AOIs). Pupil size, the proportion of fixation duration, and regression duration within and across lines were also examined. A neural network model was constructed to classify the reading practices based on these metrics.

Results: During rereading, students exhibited shorter total reading time, fixation durations, and fewer regression counts compared to first reading. Regression size was longer during rereading. Local eye movement behaviors within AOIs were also reduced. However, pupil size, the proportion of fixation duration, and regression duration within and across lines were not useful in identifying rereading. The neural network model achieved an accuracy of 0.769, precision of 0.774, recall of 0.788, and an F1-score of 0.781.

Conclusion: The findings demonstrate distinct eye movement patterns between first reading and rereading, highlighting the effectiveness of certain metrics in differentiating these practices. The neural network model provides a promising tool for rereading identification. These results expand our understanding of rereading behavior and offer valuable insights for assessing students' text familiarity.