In e-learning, it is common to expose disengaged users to various stimuli for recovering engagement. Obviously, this requires automatic assessment of users' engagement state, e.g based on certain signs of embodiment of "mind wandering" (Schooler et al., 2011). In that regard, we propose several features derived from eye blinks, which are shown to enhance perceptual decoupling, i.e. disengagement from outside stimuli in favor of internal processing (Smilek et al., 2010). As a data set enabling continuous observation of evolution of engagement, we use video recordings of users performing three tasks with different levels of user involvement: passive (viewing); semi-active (requiring listening comprehension, reasoning, inference skills) and active (requiring strategic planning, organized search, modulation of impulsive responses). The set is assessed for level of engagement by professional teachers. From the videos, we derive facial landmarks and apply real-time blink detection (Soukupova and Cech, 2016). Using polyserial correlation coefficient rho, we demonstrate that number and duration of blinks have negative correlation with engagement (rho=-0.36 and rho=-0.25, respectively), whereas normalized eye size and eye aspect ratio have a -somewhat stronger- positive correlation (rho=0.61 and rho=0.62, respectively), agreeing with (Smilek et al., 2010). This research was supported by JSPS KAKENHI Grant Number 18K18168.
Quantitative Evaluation of the Relation Between Blink Features and Apparent Task Engagement
Yucel, Z;
2019-01-01
Abstract
In e-learning, it is common to expose disengaged users to various stimuli for recovering engagement. Obviously, this requires automatic assessment of users' engagement state, e.g based on certain signs of embodiment of "mind wandering" (Schooler et al., 2011). In that regard, we propose several features derived from eye blinks, which are shown to enhance perceptual decoupling, i.e. disengagement from outside stimuli in favor of internal processing (Smilek et al., 2010). As a data set enabling continuous observation of evolution of engagement, we use video recordings of users performing three tasks with different levels of user involvement: passive (viewing); semi-active (requiring listening comprehension, reasoning, inference skills) and active (requiring strategic planning, organized search, modulation of impulsive responses). The set is assessed for level of engagement by professional teachers. From the videos, we derive facial landmarks and apply real-time blink detection (Soukupova and Cech, 2016). Using polyserial correlation coefficient rho, we demonstrate that number and duration of blinks have negative correlation with engagement (rho=-0.36 and rho=-0.25, respectively), whereas normalized eye size and eye aspect ratio have a -somewhat stronger- positive correlation (rho=0.61 and rho=0.62, respectively), agreeing with (Smilek et al., 2010). This research was supported by JSPS KAKENHI Grant Number 18K18168.File | Dimensione | Formato | |
---|---|---|---|
ECVP2019_abstract_book_05082019_p731.pdf
non disponibili
Tipologia:
Versione dell'editore
Licenza:
Copyright dell'editore
Dimensione
176.02 kB
Formato
Adobe PDF
|
176.02 kB | Adobe PDF | Visualizza/Apri |
I documenti in ARCA sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.