Combining argumentation and clustering techniques in pattern classification problems

• Autores: Gómez, Sergio Alejandro; Chesñevar, Carlos Iván
• Año de publicación: 2003
• Idioma: inglés
• Tipo de recurso: documento de conferencia
• Materia: Ciencias Informáticas; Intelligent agents; ARTIFICIAL INTELLIGENCE; Neural nets; Clustering; Machine Learning; Defeasible Argumentation; Neural networks; Fuzzy Adaptive Resonance Theory
• Institución: Universidad Nacional de La Plata

Clustering techniques can be used as a basis for classification systems in which clusters can be classified into two categories: positive and negative. Given a new instance enew, the classification algorithm is applied to determine to which cluster ci it belongs and the label of the cluster is checked. In such a setting clusters can overlap, and a new instance (or example) can be assigned to more than one cluster. In many cases, determining to which cluster this new instance actually belongs requires a qualitative analysis rather than a numerical one. In this paper we present a novel approach to solve this problem by combining defeasible argumentation and a clustering algorithm based on the Fuzzy Adaptive Resonance Theory neural network model. The proposed approach takes as input a clustering algorithm and a background theory. Given a previously unseen instance enew, it will be classified using the clustering algorithm. If a conflicting situation arises, argumentation will be used in order to consider the user’s preference criteria for classifying examples.
Eje: Agentes y Sistemas Inteligentes (ASI)
Red de Universidades con Carreras en Informática (RedUNCI)