Probabilistic reasoning on background net: An application to text categorization

Background net previously proposed is a novel approach for capturing and representing background information as a knowledge background accumulated through incremental learning on articles. As a continued study on background net, this article proposes a probabilistic reasoning on background nets by defining new acceptance measure based on conditional probabilities. Experiments on text categorization using representative data sets show that our approach, without requiring great effort in preprocessing, achieves competitive performance compared with Naive Bayes, kNN, and SVM methods.