Using support vector machines to identify literacy skills: Evidence from eye movements

Abstract

Is inferring readers' literacy skills possible by analyzing their eye movements during text reading? This study used Support Vector Machines (SVM) to analyze eye movement data from 61 undergraduate students who read a multiple-paragraph, multiple-topic expository text. Forward fixation time, first-pass rereading time, second-pass fixation time, and regression path reading time on different regions of the text were provided as features. The SVM classification algorithm assisted in distinguishing high-literacy-skilled readers from low-literacy-skilled readers with 80.3 % accuracy. Results demonstrate the effectiveness of combining eye tracking and machine learning techniques to detect readers with low literacy skills, and suggest that such approaches can be potentially used in predicting other cognitive abilities.

Hippocampal connectivity in the aftermath of acute social stress

“…Before training the classifiers, the SVM–Ranking Feature Extraction (SVM–RFE) algorithm was used to rank the features (i.e., the significantly different Granger causal values between stress and control conditions (see Fig. 4) according to their potential for discriminating between stress and control conditions (Guyon et al, 2002; Lou et al, 2017). SVM–RFE returned a ranking of the classification features (see Table 1) by training SVM with a linear kernel and removing the feature with the smallest ranking criterion.…”

“…Three aspects of the SVM models’ performance were evaluated: (1) mean classification accuracy of 30 LOSPGOCV procedures (i.e., the mean fraction of correctly classified condition out of two conditions of a participant in the test set), (2) sensitivity (i.e., the ratio of correctly classified participants in the stress condition to the total number of participants in the stress condition in the test set), and (3) specificity (i.e., the ratio of correctly classified participants in the control condition to the total number of participants in the control condition in the test set; (Akay, 2009; Lou et al, 2017).…”

“…We employed the support vector machine (SVM), a machine learning algorithm that can build a model to optimally categorize data correctly after training (Cortes and Vapnik, 1995). SVM is popular for its high flexibility and accuracy as well as its capability to deal with numerous features with few training patterns (Guyon et al, 2002; Liu et al, 2017; Lou et al, 2017).…”

What Eye Movements Reveal About Later Comprehension of Long Connected Texts

“…They trained artificial neural networks on several eye movement features computed for each slide (e.g., number of fixations, mean fixation duration, total text fixation duration, number of regressions) to generate predictions of comprehension as assessed by performance on questions presented alongside or immediately after each slide (see also Copeland, 2016;Copeland & Gedeon, 2013;Copeland, Gedeon, & Caldwell, 2015). Similarly, Mart ınez-G omez and Aizawa 2014achieved above-chance predictions of binarized (high vs. low) comprehension for short (~450 word) educational texts using a combination of linguistic features (e.g., word length) and eye movement features, with the latter being more discriminative (see also Lou, Liu, Kaakinen, & Li, 2017 for models predicting language skill).…”

What Can Eye Movements Tell us about Visual Perception Processes in Classroom Contexts? Commentary on a Special Issue

“…As stated by D'Mello et al 2020, "Such models are particularly useful when theoretical understanding is insufficient, when the data are rife with nonlinearities and interactivity, and when researchers aspire to take advantage of 'big data'." For example, Lou et al (2017) used support vector machines to identify literacy skills from readers' eye movement data, which consisted of several measures reflecting the time-course of processing of different segments of text. The models could predict students' literacy skills with high accuracy.…”