Can {AI} Mimic Human Visual Attention to Assess E-commerce Landing Page Engagement? (regular paper)

This paper presents a study on artificial intelligence (AI) models applied to extract visual saliency from images. In particular, the research assesses the accuracy of AI in replicating human-like attention mechanisms by comparing AI-generated saliency maps with eye movement data captured through eye-tracking technology. A case study is conducted to evaluate landing page engagement with viewers. Saliency maps of banners from 4 e-commerce landing pages are extracted with TranSalNet, an AI-based Visual Saliency model and compared to eye movements recorded with a webcam-based eye-tracking platform. Normalised Scanpath Saliency (NSS), Kullback-Leibler Divergence (KL-Div), and Area Under the Curve (AUC) metrics reveal AI models performing well in central regions of visual stimuli while exhibiting some false positives and false negatives in peripheral areas. The study offers insights into visual attention and e-commerce landing page assessment from a computational viewpoint