Text categorization, which consists of automatically assigning documents to a set of categories, usually involves the management of a huge number of features. Most of them are irrelevant and others introduce noise which could mislead the classifiers. Thus, feature reduction is often performed in order to increase the efficiency and effectiveness of the classification. In this paper, we propose to select relevant features by means of a family of linear filtering measures which are simpler than the usual measures applied for this purpose. We carry out experiments over two different corpora and find that the proposed measures perform better than the existing ones.

The efficiency and effectiveness of the retrieval of documents which are relevant to a certain topic or user query can be improved by means of the clustering of similar documents as well as by introducing parallel strategies. In this paper we explore the use of unsupervised learning, using clustering algorithms based on neural networks, as well as the introduction of NOW Architectures, a kind of low-cost parallel architecture, and study the impact on Information Retrieval.

In our first participation in the Cross Language Evalution Forum, we tested the performance of a clustering method which uses self organizing maps.

Decision tree induction and rule production methods have been proven as efficient tools in concept learning or data mining tasks. These approaches exhibit a good performance even when there are cognitive uncertainties in the data. Most systems in this paradigm use the information gain criterion an selecting attributes when learning. This paper presents an alternative approach. A heuristic measure of the impurity level of rules when dealing with fuzzy data is described and used in a classification algorithm. Results on the Sports classification problem are reported and compared with those of other learning algorithms.

espanolLos sistemas de aprendizaje automatico, a partir de una coleccion de ejemplos de entrenamiento, deben concluir unos mecanismos de clasificacion fiables sobre los casos no vistos. Dado que el tamano del espacio de busqueda es enorme, se deben usar metodos heuristicos. En este articulo presentamos una nueva medida heuristica de la impureza de las reglas que puede ser usada en una gran variedad de tareas en el contexto del aprendizaje automatico. De hecho, ha sido usada con exito en nuestro sistema ABANICO. Para finalizar este articulo presentamos algunos resultados experimentales en problemas con y sin ruido. EnglishStarting from a collection of training examples, machine learning systems must find reliable classification mechanisms for unseen cases. Since the searching space has a huge size, heuristic methods are used. In this paper we present a new heuristic measure for the impurity of rules that can be used in a variety of tasks in a machine learning environment. In fact, it has been succesfully used in our system ABANICO. To close the paper we report some experimental results in problems with and without noise.