Scalability of Text Classification

We explore scalability issues of the text classification problem where using (multi)labeled training documents we try to build classifiers that assign documents into classes permitting classification in multiple classes. A new class of classification problems, called ‘scalable’ is introduced that models many problems from the area of Web mining. The property of scalability is defined as the ability of a classifier to adjust classification results on a ‘per-user’ basis. Furthermore, we investigate on different ways to interpret personalization of classification results by analyzing well known text datasets and exploring existent classifiers. We present solutions for the scalable classification problem based on standard classification techniques and present an algorithm that relies on the semantic analysis using document decomposition into its sentences. Experimental results concerning the scalability property and the performance of these algorithms are provided using the 20newsgroup dataset and a dataset consisting of web news.