Text mining of wine reviews to investigate quality markers of ‘Nebbiolo’ wines from Valtellina
In Valtellina zone (north Italy), the winemaking of ‘Nebbiolo’ grapes leads to the production of two main wine types: classic red wines from fresh grapes, usually classified as Valtellina Superiore DOCG (mandatory oak aging) or Rosso di Valtellina DOC, and the Sforzato di Valtellina DOCG, which is produced using withered grapes according to traditional product specification and subjected to mandatory oak aging process. The withering process influences grape chemical composition and, in turn, the wine sensory profile, which is strongly linked to the wine quality and typicity perceived by consumers.In this study, text mining operations on reviews from renowned wine magazines and web sources (The Wine Advocate, Wine Spectator, James Suckling Wine Ratings, Wine Enthusiast) were used to explore the differences between wines produced using fresh (Valtellina Superiore-Rosso di Valtellina) and withered (Sforzato) ‘Nebbiolo’ grapes with the aim of characterising the sensory markers of these two peculiar products.Firstly, the similarities and differences of terms deriving from the reviews of Sforzato (132 reviews obtained) and the group Valtellina Superiore-Rosso di Valtellina (368 reviews obtained) were investigated through keyness analysis using the AntConc software, to identify the relevant keywords that can distinguish these wines. Then, the determination of sensory descriptors associated to the ratings of the reviews was performed using text mining strategies through IRaMuTeQ software. Sforzato and Valtellina Superiore-Rosso di Valtellina corpora, separately, were divided in low (≤86), medium-low (87-89), medium-high (90-92), and high (≥93) ratings and evaluated by clustering and correspondence analysis (CA), to find out the sensory attributes that can explain and are correlated to the quality and typicity of each category of Valtellina wines.The keyness analysis showed a very similar corpus between the two wine categories, given by the common variety and origin. Nevertheless, significantly different lemmas were found, with Sforzato described as more ‘rich’, with higher frequencies of ‘prune’ and ‘chocolate’ aromas, ‘robust’ and ‘full bodied’ on the palate, and with ‘velvet-like’ texture mouthfeel when compared to ‘Nebbiolo’ produced from fresh grapes. ‘Velvety’ descriptor was as well linked to high quality in Sforzato, because in CA it was correlated with the high-ratings group, as well as the ‘prune’ aroma descriptors. ‘Rich’ was specific for medium-high rating group, and ‘full’ and ‘chocolate’ for medium-low. Generally, in-mouth descriptors, such as mouthfeel and astringency, were able to discriminate the Sforzato wines according to the ranking group.This data analysis approach can be helpful in identifying key sensory descriptors for specific wine types and unravel markers of wine typicality. Furthermore, this knowledge will allow wine producers to modulate the winemaking strategies to obtain products noticeable by consumers.
Issue: IVAS 2022
1Department of Agricultural, Forest and Food Sciences (DISAFA), University of Torino
Contact the author
Text mining, Keyness analysis, Wine quality, Sensory descriptors, Wine reviews