Document Representations for Classification of Short Web-Page Descriptions

Motivated by applying Text Categorization to sorting Web search results, thispaper describes an extensive experimental study of the impact of bag-of-wordsdocument representations on the performance of five major classifiers --Naive Bayes, SVM, Voted Perceptron, kNN and C4.5. The texts represent shortWeb-page descriptions from the dmoz Open Directory Web-page ontology.Different transformations of input data: stemming, normalization,logtf and idf, together with dimensionality reduction, arefound to have a statistically significant improving or degrading effect onclassification performance measured by classical metrics -- accuracy,precision, recall, F$_1$ and F$_2$. The emphasis of the study is not ondetermining the best document representation which corresponds to eachclassifier, but rather on describing the effects of every individualtransformation on classification, together with their mutual relationships.