Böck, RonaldVenkateswaran, SiddarthNguyen, ThiDurner, DominikKlein, MaikeKrupka, DanielWinter, CorneliaWohlgemuth, Volker2023-11-292023-11-292023978-3-88579-731-9https://dl.gi.de/handle/20.500.12116/43088Winemaking and grapegrowing are sciences with a long tradition dealing with one of the most complex beverages in the world. This complexity stems from the winemaking process itself as well as the characteristics of the final product. Wine’s aroma is often described through scalar assessments, though here we are focusing on textual descriptions, transferring methods from the natural language processing (NLP) community to the wine domain, in particular to analyse the statements of human panellists. Textual descriptions were used for the classification of oaked versus unoaked wines as an initial demonstration of NLP in the wine domain. We achieved significant discrimination results of 0.79 F1-score comparing BERT and Naïve Bayes classifiers. This shows that more natural textual (and potentially spoken) descriptions of wine, being later combined with classical scalar assessments, can provide more flexibility to human panellists.enWine DescriptorsClassificationBERTNLPText/Conference Paper10.18420/inf2023_1621617-5468