Contextualizing Naive Bayes Predictions

Abstract

A classification process can be seen as a set of actions by which several objects are evaluated in order to predict the class(es) those objects belong to. In situations where transparency is a necessary condition, predictions resulting from a classification process are needed to be interpretable. In this paper, we propose a novel variant of a naive Bayes (NB) classification process that yields such interpretable predictions. In the proposed variant, augmented appraisal degrees (AADs) are used for the contextualization of the evaluations carried out to make the predictions. Since an AAD has been conceived as a mathematical representation of the connotative meaning in an experience-based evaluation, the incorporation of AADs into a NB classification process helps to put the resulting predictions in context. An illustrative example, in which the proposed version of NB classification is used for the categorization of newswire articles, shows how such contextualized predictions can favor their interpretability.