Understanding South African Chenin Blanc wine by using data mining techniques applied to published sensory data

[摘要] ENGLISH ABSTRACT: South African Chenin Blanc is the most planted grape cultivar in South Africa (SA) and is known forits versatility in wine sensory profiles. However, according to the South African wine industry,consumers are confused as to the different styles that make up Chenin Blanc wine. Currently,there are six different style classifications for South African Chenin Blanc wine that was proposedas a guideline by the Chenin Blanc Association (CBA). Previous research conducted at theUniversity of Stellenbosch was aimed at evaluating these style classifications.Previous results showed that, when using a small sample set of commercial Chenin Blanc,only two clear style categories could be identified – Fresh and Fruity and Rich and Ripe Wooded.However, there was a need to re-evaluate these style classifications using a larger sample set. Thecurrent study proposed a novel approach of using reputable and published text data on a CheninBlanc wines as a dataset as opposed to acquiring a limited amount of Chenin Blanc wines forsensory evaluation.The John Platter Wine Guide is South Africa's largest wine annual and provides a sensorydescription of each wine. Text data on Chenin Blanc was used over an investigation period ofseven years (2007-2014) as it was felt that this could give a clear indication of the sensoryproperties within the cultivar. It would also provide us with an extremely large dataset to evaluatethe current style classification of Chenin Blanc and this was one of the main aims of the study.Computed text data analysis on 2746 Chenin Blanc wines showed that there were only twoclear style categories identified – Fresh And Fruity and a Wooded style, which was conclusive withprevious research conducted. Additionally, for the first time the entire sensory spectrum of SouthAfrican Chenin Blanc was characterised and identifiable 'cues, as mentioned in the wine industry,was established. For the purpose of validation, a second white wine cultivar, Sauvignon Blanc, wasused to evaluate whether text data analysis on Chenin Blanc was conducted correctly.Furthermore, the sensorial properties that were retrieved from the dataset for thecharacterization of the full sensory spectrum were later used for the re-evaluation and validation ofthe current Chenin Blanc Aroma Wheel. The Chenin Blanc Aroma Wheel was developed in 2005and since then has never been re-evaluated. There has always been a need to re-evaluate theAroma Wheel, but there has never been a large enough dataset to do so. This then served as anovel approach as it was the first time that such a task would be attempted. A frequency count ofall the words used in the Platter Guide was used in conjunction with sensory attributes thatcurrently made up the aroma wheel. An exclusion/inclusion attribute criterion of five citations wasset. Attributes that were currently on the Aroma Wheel that were not cited at least five times wereexcluded and those that were currently not on the Aroma Wheel and were cited five times or morewere included. Additional categories and sub-categories were included to better accommodateattributes that were previously presented in the wrong category. Figure 1 illustrates the currentChenin Blanc Aroma Wheel while figure 2 illustrates the validated Aroma Wheel as proposed inthis study. Even though the two Aroma Wheels share similarities, it is evident that, for SouthAfrican Chenin Blanc, the current aroma wheel is not an accurate representative of the white winecultivar.

[发布机构] Stellenbosch University