Beyond traditional metrics: Spatial and semantic analysis of consumer emotional profiles using embeddings

Traditional analyses of emotional data obtained using the Check-All-That-Apply (CATA) method are well established for identifying frequency-based patterns but were not designed to capture semantic relationships among emotional terms. This study investigated the application of Natural Language Processing through semantic embeddings to analyze emotional data generated during beer sensory evaluations conducted with 100 consumers in an immersive Village virtual reality environment under three musical genres (Ambient, Country, and Pagode). The objective was to expand the understanding of consumers' emotional experiences during sensory evaluation. Initially, the semantic structure of the emotional term list was explored using embeddings, Uniform Manifold Approximation and Projection (UMAP), and Density-Based Spatial Clustering of Applications with Noise (DBSCAN), revealing non-random organization, lexically coherent groupings, and meaningful semantic distinctions among terms. Subsequently, participants' emotional responses were represented in a continuous vector space by synthesizing the emotions reported by each consumer. Projections using Principal Component Analysis (PCA) and UMAP enabled visualization of consumer dispersion and the identification of distinct emotional profiles across the musical contexts. Clustering analyses revealed intra-context emotional heterogeneity, with profiles characterized by both positive and lower-valence emotions. In some cases, these profiles were not directly associated with hedonic scores, indicating that product acceptance was not always linearly related to reported emotional valence. Furthermore, the embedding-based approach captured subtle emotional nuances and ambiguities in self-reported responses that would likely remain undetected using traditional analyses. These findings demonstrate that semantic modeling expands the interpretative potential of CATA emotional data by incorporating semantic relationships and enabling a more integrated understanding of consumer responses.