Key Points
- Researchers from the University of Copenhagen used gas chromatography-mass spectrometry (GC-MS) along with Fourier transform infrared (FT-IR) spectroscopy to analyze 89 German white wine samples. By applying machine learning algorithms to these instrumental data, they demonstrated that it is possible to predict Vivino user ratings, offering a data-driven method to estimate perceived wine quality.
- ·Vivino's crowd-sourced ratings exhibit central tendency bias but collectively reflect a reliable consensus on wine quality. Studies show that Vivino users often evaluate wines similarly to professionals and can even reveal regional or weather-based patterns in quality perception, underscoring the platform’s value in consumer research.
- This study not only highlights the predictive power of instrumental techniques in assessing consumer preferences but also suggests a roadmap for integrating user reviews, sensory analysis, and expert input. This multidisciplinary approach could evolve into a comprehensive model for evaluating and marketing wine based on its chemical and sensory profile.
Researchers at the University of Copenhagen (Denmark) aimed to conduct a feasibility study to predict Vivino quality ratings, an online platform designed to allow users to rate and describe various wines. The research team used a combination of instrumental platforms and machine learning (ML) algorithms in their analysis. The volatile organic compound (VOC) compositions of 89 commercial German white wine samples were obtained by dynamic headspace extraction (DHE) gas chromatography-mass spectrometry (GC–MS), while, simultaneously, 18 physicochemical parameters were determined using Fourier transform-infrared (FT-IR) spectroscopy. A paper based on their work was published in Food Chemistry (1).
Designed for wine enthusiasts, Vivino allows its users to rate and describe wines on a scale of 1–5, providing an average user rating for each wine surveyed (2). A quality regarding this rating system is the uncommon use of extreme ratings, as users tend to rate the wines near the median, which results in a normal distribution of the scores when the sample size is sufficiently large. This tendency, often referred to as central tendency bias, is well documented in various domains which utilize rating or measurement systems (3). Vivino, on the other hand, leverages more of a “wisdom of the crowd” concept, suggesting that the average of multiple independent estimates is generally more accurate than any single estimate (4,5).
This crowd-sourced data provides a unique viewpoint concerning wine quality. Through the platform, visitors can discover wines new to them, access opinions from fellow enthusiasts, and contribute to a growing database of wine knowledge, thus making wine selection more accessible and informed for everyone (1).
The combination of GC–MS and FT–IR produces information on the wine’s volatile profile, which is crucial for aroma attributes, and the chemical constituents, which are vital for its taste, according to the researchers. These methods have been previously used in combination with chemometrics and ML methods for the identification of key sensory and chemical attributes which drive consumer liking (8); the modelling of quality traits and the making of predictions regarding sensory profiles of Australian Pinot Noir (9); the investigation of Cabernet Sauvignon grape parameters which predict the sensory properties of the finished wine (10); and the characterization of Danish Solaris white wines (11).
Vivino ratings have been used in several studies to investigate consumer perception of wine. For example, one investigation found that Vivino users demonstrate rich knowledge of wine similar to experts, without being heavily influenced by price (6); a second study analyzing regional wine ratings which suggested a community effect on perceived quality, referred to a particular wines popularity within the Vivino community (7); a third found a correlation between Vivino ratings and professional critics, and reported that both respond to weather changes (5).
The researchers are confident that their results show the possibility to predict consumer quality ratings from chemical parameters, providing both a framework for estimating product quality and deeper insight into the chemical attributes that influence consumer perception of wine quality. Their findings confirm the value of crowd-sourced data in the exploration of the relationship between chemical compositions and consumer acceptance, as they believe that this data captures the perspective of the average consumer. This approach lays the foundation for future investigations into predicting consumer preferences and perceived quality directly from analytical data. Furthermore, the potential of expanding this model with additional modalities—such as instrumental techniques, text-based processing of user reviews, sensory analysis, and expert ratings—may provide a comprehensive tool for the assessment of wine quality (1).
References
- Hjertholm, F.; Goetz, R.; Schneide, P. A. et al. Uncorking White Wine Liking: Combining Analytical Chemistry and Chemometrics with Crowd-Sourced Data to Predict Quality Ratings. Food Chem. 2025, 492 (Pt 1), 145376. DOI: DOI: 10.1016/j.foodchem.2025.145376
- The Vivino Wine Rating System: Credibility of the Crowd. Vivino website. https://www.vivino.com/wine-news/vivino-5-star-rating-system (accessed 2025-07-09)
- Akbari, K.; Eigruber, M.; Vetschera, R. Risk Attitudes: The Central Tendency Bias. EJDP 2024, 12, 100042. DOI: 10.1016/j.ejdp.2023.100042
- Fiechter, J. L.; Kornell, N. How the Wisdom of Crowds, and of the Crowd Within, Are Affected by Expertise. CRPI 2021, 6, 1-7. DOI: 10.1186/s41235-021-00273-6
- Kopsacheilis, O.; Analytis, P. P.; Kaushik, K.; Herzog, S. M. et al. Crowdsourcing the Assessment of Wine Quality: Vivino Ratings, Professional Critics, and the Weather. J. Wine Econ. 2024, 19 (3), 285-304. DOI: 10.1017/jwe.2024.20
- Kotonya, N.; De Cristofaro, P.; De Cristofaro, E. Of Wines and Reviews: Measuring and Modeling the Vivino Wine Social Network. In 2018 IEEE/ACM International Conference on Advances in Social Networks Analysis and Mining (ASONAM); IEEE, 2018, 387-392. DOI: 10.48550/arXiv.1804.10982
- Gastaldello, G.; Schäufele-Elbers, I.; Schamel, G. Factors Influencing Wine Ratings in an Online Wine Community: The Case of Trentino–Alto Adige. J. Wine Econ. 2024, 19 (1), 19-40. DOI: 10.1017/jwe.2024.2
- Dahal, K. R.; Dahal, J. N.; Banjade, H. et al. Prediction of Wine Quality Using Machine Learning algorithms. Open J. Stat. 2021, 11 (2), 278-289. DOI: 10.4236/ojs.2021.112015
- Fuentes, S.; Torrico, D. D.; Tongson, E. et al. Machine Learning Modeling of Wine Sensory Profiles and Color of Vertical Vintages of Pinot Noir Based on Chemical Fingerprinting, Weather and Management Data. Sensors 2020, 20 (13), 3618. DOI: 10.3390/s20133618
- Niimi, J.; Tomic, O.; Næs, T. et al. Objective Measures of Grape Quality: From Cabernet Sauvignon Grape Composition to Wine Sensory Characteristics. LWT 2020, 123, 109105. DOI: 10.1016/j.lwt.2020.109105
- Liu, J.; Toldam-Andersen, T. B.; Petersen, M. A. et al. Instrumental and Sensory Characterisation of Solaris White Wines in Denmark. Food Chem. 2015, 166, 133-142. DOI: 10.1016/j.foodchem.2014.05.148
