Developing a Psychometric Tool to Measure the Emotional Impact of Visual Content

Abstract

This thesis investigates the human valence response to sequences of visual images. We f irst use crowd-sourcing and a novel nine-point psychometric scale to estimate human valence responses to individual images from the OASIS image set with high reliability (split-half Spearman rank-correlation ρ = 0.95). In a separate group of human participants, we then estimate valence responses following short, random sequences of those images (of length ≤ 10). Our key finding is that these sequence-contingent valence responses can be closely predicted by a simple linear combination of the estimated human valence responses to individual images (held-out ρ = 0.94). The combination weights are largest for the final image in the sequence; intuitively, this means the final image by itself can make predictions with high goodness-of-fit (ρ = 0.87). In summary, this research shows new evidence for a simple relationship between valence responses to individual images and valence responses to image sequences, with implications for future studies and practical applications in psychological assessment and beyond.