Attention Prediction on Webpage Images using Multilabel Classification

Image elements (called webpage images) are prominent in drawing the user attention on webpages. Quantitatively predicting the attention on webpage images help in their synthesis as well as rendering. This paper presents the attention prediction as a multilabel classification problem for the webpage images. Firstly, fixated images were assigned multiple quantified visual attention (called fixation-indices) based on users’ sequential attention allocation on webpages. Subsequently, fixated images’ visual features such as intensity, brightness, color histograms, position and size on respective webpages, were extracted. A multilabel classification problem formulated using the visual features and fixation-indices was solved using multiple binary support vector machine (binary-SVM) classifiers. The performance of the proposed approach was analyzed through a free-viewing eye-tracking experiment conducted on 36 real-world webpages with 42 participants. Our model outperforms the baseline with precision, recall, Fl-score, and an accuracy of 84.14%, 78.91%, 76.57%, 67.49% respectively with a hamming loss of 23.62% and a subset 0/1 loss of 66.50%.