Machine Learning for Real-Time Analysis of Reader Attention by Facial Expression and Eye Tracking Data.

Mangili, Francesca and Antonucci, Alessandro and Poulopoulou, Maria Fani and Werlen, Egon and Bergamin, Per (2018) Machine Learning for Real-Time Analysis of Reader Attention by Facial Expression and Eye Tracking Data. UNSPECIFIED. In: ndustrial Applied Data Science, 18-19 October 2018, Lugano. (Unpublished)

[img] Text (Short description with poster)
Mangili et al (2018).pdf - Other
Restricted to Registered users only

Download (4MB) | Request a copy


Eye tracking and facial expression recognition are nowadays well consolidated techniques for which reliable and relatively inexpensive hardware and software tools are available. A natural application of these tools is within the field of e-learning and e-education technologies. A student or an exam taker might therefore be monitored during a (e-)class or a computer-administered exam by means of these tools. As a matter of fact, eye tracking data analyses carried out in the literature and exploited in educational applications are mostly based on classical statistical techniques which do not take into account temporal aspects. Wu compares common classifiers used to infer the cognitive abilities of participants solving various visualization tasks during a controlled experiment. Classification is based on the use of statistical descriptors of eye tracking data as well as sequential patterns of user’s gaze movement. The work achieved an accuracy above 80% in the binary classification of low/high cognitive abilities using only statistical descriptors, whereas sequential pattern did not bring any improvement. Building on these results, this work aims to further explore the capability of advanced machine learning methods of exploiting the full information contained in the sequence of saccades and fixations of different areas of interest recorded by the eye tracker to improve and refine the inference about the user learning experience. Our focus is more on the degree of comprehension of the topics conveyed by the activity, which can be directly related to targeted interventions.In this study we develop a method for inferring the degree of understanding of a participant from the monitored parameters, while reading the text and assess the performance of the method on the reading experiment data. The preliminary results are encouraging as they show that machine learning techniques can be successful in extracting actionable knowledge from eye tracking data. Next step will be to assess the possibility of exploiting such information in adaptive tutoring systems.

Actions (login required)

View Item View Item