Kohonen Self-organising Maps In The Data Mining Of Wine Taster Comments
Free (open access)
125 - 139
P. Sallis, S. Shanmuganathan, L. Pavesi & M. C. J. Muñoz
Computational neural network methods are increasingly being used for researchoriented data mining tasks. Kohonen self-organising map (SOM) techniques are well established within the so-called connectionist paradigm of Artificial Intelligence where neural networks are used to extract both explicit and implicit dependency values often between data that is sometimes disparate in type and kind. The research described here seeks to elicit relationships between grape varieties and their growing conditions using SOM techniques. In addition, utilising k-means and principle component analysis (PCA) methods, the data mined and depicted by the SOM technique is shown to have dependency values that enable a clustering of terms relating to variety quality to be associated with growing condition data to produce optimal locations for each. This study is part of a larger research project that uses comparative data from New Zealand and Chile. The text mining aspect of it forms one element of a ‘toolbox’ of integrated hardware and software instruments being developed to underpin an environmental modelling methodology oriented not only towards grape growing but also generally for optimal crop production. The example described here uses data from New Zealand in the first instance. The paper begins with a summary of some historical wisdom relating to grape growing with a discussion of some previous studies and then describes the text mining of comments from wine tasters, which are statistically analysed. The results are clustered and the paper concludes with a reflection on the investigation with a pointer to future work in this aspect of the larger research project previously described. Keywords: text mining, wine taster comments and wine sensory data.
text mining, wine taster comments and wine sensory data.