Discovering the Language of Wine Reviews: A Text Mining Account

Discovering the Language of Wine Reviews: A Text Mining Account Els Lefever1, Iris Hendrickx2, Ilja Croijmans2, Antal van den Bosch2,4, Asifa Majid2,3 1Language and Translation Technology Team, Ghent University, Belgium 2Centre for Language Studies, Radboud University, Nijmegen, The Netherlands 3Max Planck Institute for Psycholinguistics, Nijmegen, The Netherlands 4Meertens Institute, Amsterdam, The Netherlands els.lefever@ugent.be, {i.hendrickx,i.croijmans,a.vandenbosch,asifa.majid}@let.ru.nl Abstract It is widely held that smells and flavors are impossible to put into words. In this paper we test this claim by seeking predictive patterns in wine reviews, which ostensibly aim to provide guides to perceptual content. Wine reviews have previously been critiqued as random and meaningless. We collected an English corpus of wine reviews with their structured metadata, and applied machine learning techniques to automatically predict the wine’s color, grape variety, and country of origin. To train the three supervised classifiers, three different information sources were incorporated: lexical bag-of-words features, domain-specific terminology features, and semantic word embedding features. In addition, using regression analysis we investigated basic review properties, i.e., review length, average word length, and their relationship to the scalar values of price and review score. Our results show that wine experts do share a common vocabulary to describe wines and they use this in a consistent way, which makes it possible to automatically predict wine character- istics based on the review text alone. This means that odors and flavors may be more expressible in language than typically acknowledged. Keywords: wine reviews, wine vocabulary, classification, supervised learning, terminology extraction 1. Introduction Few food categories are described as often as wine: a vast number of wine reviews appear in magazines, books, blogs, supermarkets, newspapers, and numerous other venues. These reviews contain descriptors, which chronicle the ap- pearance, aroma (smell), flavor, and textural attributes of wines in loving detail. This is puzzling, as previously many scholars have claimed that smells and flavors are difficult, if not impossible to put into words (Sperber, 1975; Rouby et al., 2002; Levinson and Majid, 2014). There is an ongoing debate as to whether wine is actually described in an informative manner in wine reviews. On the one hand, studies suggest wine experts use language in a consistent manner (Croijmans and Majid, 2016). Simi- larly, expert descriptions are more often correctly matched to a wine than descriptions written by novices (Solomon, 1990), with the suggestion that wine experts are more likely to use specific terminology which is more informative. On the other hand, studies suggest trained wine experts more often use vague and abstract terms (e.g., complex, attrac- tive) when describing wines (Gawel, 1997). Similarly, metaphorical language is often encountered in wine de- scriptions (Su´arez Toste, 2007; Caballero, 2007), suggest- ing wine experts employ vague and overly literary prose. Others have more directly critiqued wine reviews as be- ing uninformative and mere “purple prose” (Quandt, 2007) (p.130). Here, we study how wine experts express properties of wines in their reviews. Wine reviews convey both sensory descriptions of wines, as well as objective properties such as color, grape type, country of origin, …

Content truncated to 500 words