A KNOWLEDGE-BASED MODEL OF CONTEXT-DEPENDENT ATTRIBUTE PREFERENCES FOR FAST MOVING CONSUMER-GOODS

Modelling techniques which integrate statistical and knowledge-based m ethods are shown in this paper to provide the basis for a richer disco urse about the nature of markets than either statistical methods or ve rbal analyses alone. This result is demonstrated in relation to a mode l in which consumer preferences are dependent upon the activity for wh ich purchases are made. This knowledge-base of the model incorporates qualitative judgements of domain experts, The model is then tested aga inst EPOS data on the sales to test the consistency of these judgement s. We then exhibit an application of this model to the UK market for c ertain alcoholic beverages. We show that the context-dependent model i s at least as robust as models obtained by conventional statistical an alysis. The context-dependent model needs far Less data for effective parameterisation and contains more information that is directly useful to domain experts. However, the statistical model is less computation ally intensive. (C) 1997 Elsevier Science Ltd.