Computação baseada em casos: contribuições metodológicas aos modelos de indexação, avaliação, ranking, e similaridade de casos.

Abstract

Perceptiun provides agents with information about the word they inhabit. To perceive by using similarities and analogies is one of the most advanced forms of perceiving. It constitutes one of the most fundamental aspects of human cognition. Computationally, similarity is the comer-stone upon which relies the entire case-based reasoning technology (CBR), as an intelligent computing paradigm. In its origin and nature, the CBR technology is affected by similarity in ali over its operational aspects. The thesis here detailed has the general objective of exploring the CBR computational paradigm regarding to the similarity point of view and its implication for these paradigm processes. The start point, is the recognition of insufficiencies of current similarity approaches, in general, andparticularly, in the CBR domain where the CBR processes are dominated by euclidean metrics and statistical nearest-neighbours techniques. In this thesis, we propose to enhance the CBR methodologies by developing the application of Tversky-Gati cogniíive íheory to the CBR methodologies of case indexing, case evaluation, case similarity itself, cascranking, and case-based query-answering. We experiment with this computational extent of the cognitive theory in the empirical domain oíloanunderwriting where a credit underwriter has to decide on recommending or rejecting credit applications based on their attribute similarities.