Analysis & Experimental Results ∗

In recent years we have seen a tremendous growth in the volume of text documents available on the Internet, digital libraries, news sources, and company-wide intranets. Automatic text categorization, which is the task of assigning text documents to pre-specified classes (topics or themes) of documents, is an important task that can help both in organizing as well as in finding information on these huge resources. Text categorization presents unique challenges due to the large number of attributes present in the data set, large number of training samples, and attribute dependencies. In this paper we focus on a simple linear-time centroid-based document classification algorithm, that despite its simplicity and robust performance, has not been extensively studied and analyzed. Our extensive experiments show that this centroid-based classifier consistently and substantially outperforms other algorithms such as Naive Bayesian, k-nearest-neighbors, and C4.5, on a wide range of datasets. Our analysis shows that the similarity measure used by the centroid-based scheme allows it to classify a new document based on how closely its behavior matches the behavior of the documents belonging to different classes, as measured by the average similarity between the documents. This matching allows it to dynamically adjust for classes with different densities. Furthermore, our analysis shows that the similarity measure of the centroid-based scheme accounts for dependencies between the terms in the different classes. We believe that this feature is the reason why it consistently outperforms other classifiers that cannot take these dependencies into account.