Uncorking white wine liking: Combining analytical chemistry and chemometrics with crowd-sourced data to predict quality ratings

Quality perception is a crucial aspect of wine research, particularly regarding consumer acceptance and quality impression. These assessments are challenging because consumers frequently incorporate extrinsic cues when evaluating hedonic quality. A unique perspective on wine quality is provided by the platform Vivino, which utilizes crowd-sourced consumer reviews on a scale of 1 to 5. This study investigates the relationship between chemical composition and consumer liking, using Vivino ratings as quality endpoints. German white wines were analysed using gas chromatography-mass spectrometry and Fourier transform infrared spectroscopy to establish comprehensive chemical profiles. Mid-level data fusion was applied, followed by variable selection to identify significant chemical features used to construct a Partial Least Squares regression model for predicting Vivino quality ratings. Our findings demonstrate that chemical properties can predict consumer-perceived quality parameters, validating our hypothesis and underscoring the potential for further research to refine and expand predictive models in wine quality assessment.