Predicting reading comprehension scores from eye movements using artificial neural networks and fuzzy output error

Abstract

Predicting reading comprehension from eye gaze data is a difficult task. We investigate the use of artificial neural networks (ANNs) to predict reading comprehension scores from eye gaze collected from participants who read and completed an online tutorial in our lab. Problems such as large feature sets and small highly imbalanced data sets compound to make this task even more complex. We propose using fuzzy output error (FOE) as an alternative performance function to mean square error (MSE) for training feed-forward neural networks to overcome these problems. We show that the use of FOE as the performance function for training ANNs provides significantly better classification of eye movements to reading comprehension scores. ANNs with three hidden layers of neurons gave the best classification results especially when FOE is used as the performance function for training. In these cases we found up to 50% reduction in misclassification rates compared to using MSE. We found that ANNs give optimal classification results in comparison to other classification techniques. When FOE is used as the performance function for training the ANNs the misclassification rates are halved compared to the other techniques. Cluster analysis was performed on one of the more complex data sets. Interesting reading behaviour properties were found within the data set. The intended use of this research is in the design of adaptive online learning environments that use eye gaze to predict user comprehension from reading behavior.

What Eye Movements Reveal About Later Comprehension of Long Connected Texts

“…Regression fixations were any fixations that were on a word with an index lower than that of the previous word that was fixated on. These features have been used in similar modeling efforts (Copeland, 2016;Copeland & Gedeon, 2013;Copeland et al, 2014Copeland et al, , 2015Mart ınez-G omez & Aizawa, 2014) and have been empirically linked to factors that influence text comprehension, including mind-wandering (e.g., Reichle et al, 2010;Uzzaman & Joordens, 2011) and text difficulty (Rayner et al, 2006). Although it is not an eye movement feature, we also included page reading time in seconds as it is perhaps the most ubiquitous dependent measure in reading research and has been linked to comprehension (e.g., Mills, Graesser, Risko, & D'Mello, 2017).…”

A Discriminative Model for Identifying Readers and Assessing Text Comprehension from Eye Movements

“…Existing attempts to exploit this eye-mind connection and actually use a reader's eye movements to predict text comprehension have crucial limitations. Copeland et al [20,21,22] use the saccades between a comprehension question and the text as a feature to predict the response accuracy on this very question. Hence, these models are not trained to infer reading comprehension from the eye movements while reading a text, as claimed by the authors, but rather predict response accuracy on a question from the answer-seeking eye movements of the user.…”

Near Real-Time Comprehension Classification with Artificial Neural Networks: Decoding e-Learner Non-Verbal Behavior

“…Analysing objective learner comprehension of on-screen information has been approached using eye-tracking techniques and body attached sensors [9], [10], [11], [21], [22]. The literature shows positive results in providing real-time comprehension classifications for mental processing of onscreen materials in a variety of laboratory experiments.…”

“…Literature on classifying learner cognitive states from NVB presents a broad range of technical approaches and assessment contexts. Related work (section 2) discusses comprehension assessment in dyadic verbal information recall tasks [7], [8], use of specialised high-speed eye-tracking cameras [9], [10] and heat-maps [11] to classify reading comprehension, classifying learning activity states using facial expressions [12] and predicting self-reported task difficulty by coarse head movements [13]. Despite the rich literature in the field, a generalised, cost-effective, practical and nonintrusive method of objectively classifying e-learner comprehension of on-screen information is absent.…”