Continuous Prediction of Web User Visual Attention on Short Span Windows Based on Gaze Data Analytics.

Area of Science:

• Human-Computer Interaction
• Computer Vision
• Web Science

Background:

• Understanding user visual attention is crucial for improving web browsing experiences and adapting dynamic interfaces.
• Current methods often rely on static website analysis, which is insufficient for increasingly interactive and dynamic web content.

Purpose of the Study:

• To propose and demonstrate a novel method for predicting user visual attention on specific regions of websites with dynamic components.
• To develop a system that predicts visual attention without needing constant updates of the current website layout.

Main Methods:

• Introduced the concept of 'visit intention' to predict future user gaze fixation probabilities.
• Utilized eye-tracking data from a population browsing a specific website to train personalized prediction models.
• Employed multilabel classification models incorporating individual visual kinetics features.

Main Results:

• Achieved an average area under the curve (AUC) of 84.3% and an average accuracy of 79% in predicting visual attention.
• Demonstrated the effectiveness of the prediction method even with a small user group.
• Consistently identified user visual kinetics features as significant predictors across cross-validation evaluations.

Conclusions:

• The proposed method effectively predicts user visual attention on dynamic web regions using historical data and individual characteristics.
• Personalized prediction models leveraging visual kinetics show promise for enhancing user experience on complex websites.