Cluster Analysis in Science and Technology: An Application in Research Group Evaluation
Alexandre Leopoldo Gonçalves, Roberto Carlos dos Santos Pacheco, Aran Bey Tcholakian Morales, Viníus Medina Kern · The Florida AI Research Society · 2002
Parametric methods for science and technology evaluation are frequently rejected because evaluators prefer a subjective approach, usually achieved through peer review analysis. This paper takes a complementary approach to classify research activities by means of machine learning systems. We propose the use of a non-supervised neural network in the building of a ranking of Brazilian research groups. Two indexes are built, expressing productivity and qualification of research groups. The indexes and their relationship are used in the classification of research groups in five categories ( strata). The results have been consistent with a parametric algorithm currently used by the Brazilian National Research Council (CNPq). In conclusion we suggest the plausibility of applying machine learning in knowledge extraction from science and technology databases.