Brand Positioning Maps: Latent Allocation Model versus Correspondence Analysis
(A2023-113599)

AUTHORS

pablo marshall, puc

ABSTRACT

Correspondence analysis is a methodology that reduces the dimensions of a contingency table to enable visualizing, in a perceptions map, the relationship between brands of a category and certain attributes or concepts. In this context, contingency tables have the form of word frequency, similar to the frequency matrices used in text analytics. This study proposes the use of topic modelling, particularly the Latent Dirichlet Allocation family of models, as an alternative methodology to reduce the dimension of a contingency table and represent brands and attributes or concepts in a perception map. In relation to correspondence analysis, the proposed methodology allows a simpler interpretation of the latent dimensions, uses ternary maps to represent three dimensions in a two-dimensional plane, and is based on a model that permits predicting and tracking brand perceptions. A simulation exercise and an application show the gains from using the proposed methodology.