The Induction of Consumer Preference Models using Evolutionary Computation

A technique is described whereby context-dependent consumer preference models can be automatically induced using algorithms taken from the field of evolutionary computation. In order to make this possible an abstract meta-model is constructed to relate classes of preference models to aggregate price and sales data. This meta-model is broadly consistent with existing consumer preference models, but its prime purpose is to provide an appropriate framework for the technique. The technique proceeds as following: a marketing practitioner specifies the relevant market attributes and the perceived values of these attributes for each product; the algorithm then induces models within this framework that explain the aggregate data. We tested this technique on markets for alcoholic beverage. We found good fits with the data using relatively short sequences of in-sample data, but more importantly it gave qualitative information about the possible contexts of consumer purchases.