Análisis de Datos de Eye-Tracker durante textos digitales.
Loading...
Date
2025
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Universidad de Concepción
Abstract
Esta memoria presenta un estudio basado en un proyecto centrado en el análisis de la lectura digital mediante el uso de tecnología de seguimiento ocular (Eye-Tracking). Se evaluó el comportamiento visual de 41 participantes universitarios, previamente clasificados como lectores hábiles o menos hábiles en comprensión lectora, mientras leían textos digitales estructurados jerárquicamente o en red, y con niveles de dificultad normal o difícil.
Las métricas extraídas incluyeron duración promedio de fijaciones, longitud de sacadas, número de f ijaciones por minuto, cantidad de regresiones y tiempo total de lectura. Estas fueron analizadas median te pruebas estadísticas (t-test, ANOVA bifactorial y modelos lineales mixtos), obteniendo diferencias significativas entre los tipos de lectores en varias de estas métricas. En particular, los lectores hábiles mostraron patrones más eficientes, como fijaciones más cortas y sacadas más largas. Respecto a la difi cultad del texto, en las lecturas difíciles aumentaron el tiempo de lectura y la cantidad de fijaciones. No hubo una interacción significativa entre el tipo de lector y la dificultad del texto.
Además, como complemento, se implementaron modelos de Machine Learning y Redes Neuronales para clasificar a los participantes según su tipo de lector, a partir de movimientos oculares durante una lectura. Se entrenaron múltiples modelos (Random Forest, SVC, XGBoost, MLP, entre otros), optimiza dos con búsqueda bayesiana mediante Optuna y validados con validación cruzada estratificada. El mejor rendimiento se obtuvo con XGBoost, alcanzando un F1-score de 0.753.
Los resultados confirman que las métricas de Eye-Tracking reflejan diferencias lectoras significativas entre lectores hábiles y de bajo desempeño, y que estas métricas pueden ser utilizadas para clasificar a los participantes según su capacidad lectora en entornos digitales.
This thesis presents a study focused on digital reading analysis using Eye-Tracking technology. The visual behavior of 41 university students was evaluated, previously classified as either skilled or low performing readers in reading comprehension, while they read digital texts structured hierarchically or as networks, and with normal or difficult levels of complexity. The extracted metrics included average fixation duration, saccade length, fixations per minute, num ber of regressions, and total reading time. These were analyzed using statistical tests (t-test, two-way ANOVA, and mixed linear models), revealing significant differences between reader types for several of these measures. Specifically, skilled readers showed more efficient patterns, such as shorter fixations and longer saccades. Difficult texts increased reading time and the number of fixations. No significant interaction was found between reader type and text difficulty. In addition, Machine Learning and Neural Network models were implemented to classify participants based on their reading performance using eye movement data. Various models were trained (Random Fo rest, SVC, XGBoost, MLP, among others), optimized using Bayesian search with Optuna, and evaluated with stratified cross-validation. The best performance was achieved with XGBoost, reaching an F1-score of 0.753. Theresults confirm that Eye-Tracking metrics reveal meaningful differences between skilled and low performing readers and can be used to classify individuals according to their reading ability in digital environments.
This thesis presents a study focused on digital reading analysis using Eye-Tracking technology. The visual behavior of 41 university students was evaluated, previously classified as either skilled or low performing readers in reading comprehension, while they read digital texts structured hierarchically or as networks, and with normal or difficult levels of complexity. The extracted metrics included average fixation duration, saccade length, fixations per minute, num ber of regressions, and total reading time. These were analyzed using statistical tests (t-test, two-way ANOVA, and mixed linear models), revealing significant differences between reader types for several of these measures. Specifically, skilled readers showed more efficient patterns, such as shorter fixations and longer saccades. Difficult texts increased reading time and the number of fixations. No significant interaction was found between reader type and text difficulty. In addition, Machine Learning and Neural Network models were implemented to classify participants based on their reading performance using eye movement data. Various models were trained (Random Fo rest, SVC, XGBoost, MLP, among others), optimized using Bayesian search with Optuna, and evaluated with stratified cross-validation. The best performance was achieved with XGBoost, reaching an F1-score of 0.753. Theresults confirm that Eye-Tracking metrics reveal meaningful differences between skilled and low performing readers and can be used to classify individuals according to their reading ability in digital environments.
Description
Tesis presentada para optar al título de Ingeniero/a Civil Biomédico.
Keywords
Registro visual, Eye tracking, Lectura, Medios digitales