Veebilehtede hindamine tehisintellekti abil

Abstract

This bachelor’s thesis investigates how accurately different artificial intelligence (AI) models can assess the visual aesthetics of websites compared to human judgments. The study analyzed four AI models – GPT-4o, Grok-2 Vision, Google Cloud Vision, and SAP – using two types of evaluation prompts: a short general score and a detailed score based on VisAWI criteria. The results showed that GPT-4o, when given a detailed prompt, most closely matched human evaluations. Model performance was evaluated using correlation, AUROC, and RMSE metrics. In addition, the stability and criticality of the models were analyzed. The study contributes to the field of automated aesthetic quality assessment and provides practical insights for future applications in design value analysis.