When Naïve is not Enough: Bringing Naïve Bayes Text Categorization to "Surface"

Since information has become more and more available in digital format, especially on the World Wide Web, organizing and classifying digital documents, making them accessible and presenting them in a proper way are becoming important issues. Digital Library Management Systems (DLMSs) are an example of systems that manage collections of multi-media digitalized data and include components that perform the storage, access, retrieval, and analysis of the collections of data. Solutions for content organization, access and interaction are required in order to let users express their information needs, especially when the specific request is not clear in their mind [2]. Automatic categorization systems may be used to build classification schemes (or taxonomies, or subject hierarchies), in order to browse, explore, and retrieve resources from collections of digital objects. Visualization frameworks that give straightforward graphical explanation of organization and classification of data have been successfully designed and implemented [3]. However, this visualization approaches have been rarely, if never, applied in the area of textual document categorization. However, these visualization approaches have been rarely applied in the area of textual document categorization. In fact, the representation of textual information is particularly challenging: how can the semantics of textual documents be captured and represented through graphs? A recent probabilistic approach on Automated Text Categorization (ATC) that represent documents on the two– dimensional space [1, 4] has shown to be a valid visualization tool to understand the relationships between categories of textual documents, and help users to visually audit the classifier and identify suspicious training data. In this work, we apply the same idea of the twodimensional representation of documents to the case of the Naı̈ve Bayes (NB) classifier. 2 Naı̈ve Bayes Categorization