Exploiting Negative Categories and Wikipedia Structures for Document Classification

摘要

This paper explores the effect of profile based method for classification of Wikipedia XML documents. Our approach builds two profiles, exploiting the whole content, Initial Descriptions and links in the Wikipedia documents. For building profiles we use the negative category information which has shown to perform well for classifying unstructured texts. The performance of Cosine and Fractional Similarity metrics is also compared. The use of two classifiers and their weighted average improves the classification performance.