SciELO - Scientific Electronic Library Online

vol.6 número1Efeitos do nível de benefício utilitário sobre a duração do comportamento de procura por produtos índice de autoresíndice de assuntospesquisa de artigos
Home Pagelista alfabética de periódicos  

Revista Psicologia Organizações e Trabalho

versão On-line ISSN 1984-6657


MARTINS, Weber; NALINI, Laura Eugênio Guimarães  e  TSUKAHARA, Fernando Pirkel. Context-sensitive multidimensional ranking: an alternative technique to data complexity. Rev. Psicol., Organ. Trab. [online]. 2006, vol.6, n.1, pp. 265-293. ISSN 1984-6657.

Many applications, as comparison among products represented by a large number of attributes, require ordering of instances represented by high dimensional vectors. Despite the reasonable quantity of papers on classification and clustering, papers on multidimensional ranking are rare. This paper expands a generic neurogenetic ranking procedure based on one-dimensional Self-Organizing Maps (SOMs). Their typical similarity metric is modified to a weighted Euclidean metric and automatically adjusted by a genetic algorithm, a heuristic search (optimization) technique. The search goal is the best ranking that matches the desired probability distribution (provided by experts) leading to a context-sensitive metric. In order to ease expert agreement, the technique relies on consensus about the best and worst instances only. In addition to providing a ranking, the derived metric is also useful for reducing the number of dimensions (questionnaire items in some situations) and for modeling the data source. In practical terms, a technique to convert subjective knowledge into objective scores is presented, creating a specific and operational model capable to deal with new situations. This technique is exemplified by two cases: ranking of data from blood bank inspections and client segmentation in agribusiness. On the theoretical point ofview, instead, the proposed system has presented a way to stabilize results from SOMs by imposing expert constraints, leading to context-sensitive multidimensional ranking. Despite the fact that SOMs are a class of artificial neural networks, they are radically different from the neural model usually employed in Business and Economics studies, the multilayer perceptron with backpropagation training algorithm. The main objective of this article is, therefore, to present a powerful combination of techniques originated in Artificial Intelligence - a multidisciplinary field more related to Engineering than to Mathematics, where Statistics has its origins and deductive basis

Palavras-chave : ranking; Self-Organizing Map; Genetic Algorithm; multidimensionality; data reduction.

        · resumo em Português     · texto em Inglês     · Inglês ( pdf )


Creative Commons License