Upload
others
View
0
Download
0
Embed Size (px)
Citation preview
IdentificationofuserconfusioninawebapplicationAuthor:MichalHuckoSupervisor: prof.MáriaBieliková
”Confusionisasituationinwhichpeopleareuncertainaboutwhattodoorareunabletounderstandsomethingclearly.”
Histogram representation
of features
Mouse movements
Extracted features
velocity, acceleration, distance,
duration, jerk, ...
Confusion
classifier
If user confused,
show guides
Predict user confusion in real time for webapplications based on logged mouse interactiondata
Goal
• 6 tasks at FiroTour travel agency’s portal• 59 participants• confusionwasloggedwithasoftwarebutton
(exclamationmark)situatedintherighttopcornerofthescreen
• 95 out of 354 tasks with button press
Eyetrackinguserstudy
Results• 99% precision on not confused data (real time)• 40% precision on confused data (real time)• 94% precisiononnotconfuseddata
(predictionaftersession)• 40% precisiononconfuseddata
(predictionaftersession)Wecomparedlogisticregression,randomforestandmultilayerperceptron classifier.
Our method is part of an ongoing commercial projectYesElf. We implemented a production ready module ableto predict the confusion in real time. Hundreds of YesElf’scustomers can use our solution daily.
YesElf
Mousefeatures• Horizontal velocity• Vertical velocity• Velocity• Acceleration• Distance
• Jerk• Num. of movements