Predicting the Reader's English Level From Reading Fixation Patterns Using the Siamese Convolutional Neural Network.

Area of Science:

• Cognitive Science
• Computer Science
• Education

Background:

• Reading comprehension is linked to cognitive patterns and metacognitive processes like comprehension monitoring.
• Individual cognitive patterns during reading can potentially predict reading comprehension levels.

Purpose of the Study:

• To investigate if reading cognitive patterns, captured via eye-tracking, can predict English reading comprehension levels.
• To develop and evaluate deep learning models for predicting reading proficiency.

Main Methods:

• 80 college students were divided into pass and non-pass groups based on College English Test Band Six (CET-6) criteria.
• Eye-tracking heatmaps of fixation counts were collected during reading comprehension tests.
• Siamese convolutional neural network models were trained and fine-tuned to predict English levels, with predictions integrated using Soft Voting.

Main Results:

• A Siamese network model with a cluster radius of 25 pixels and L1 norm distance achieved superior performance.
• The Area Under the Curve (AUC) values for the Siamese networks reached 0.941 (trained from scratch) and 0.956 (pre-trained with fine-tuning).

Conclusions:

• Individual reading cognitive patterns, as measured by eye-tracking, are predictive of reading comprehension levels.
• Deep learning models, particularly Siamese networks, demonstrate high accuracy in predicting English proficiency based on eye-movement data.