EFFECTS OF TEXT REPRESENTATION METHODS IN SEMANTIC SPACE ON CLASSIFYING PERFORMANCE

Abstract

The most discussed issue about classification of texts is how to represent them. Words, stem words, character ngrams and semantic spaces are the common methods. In this research latent semantic indexing and semantic space based on co-occurrence matrix methods are compared with other methods on 30 classed data set. According to other methods the semantic space based on co-occurrence matrix method performs higher success. In addition, performance success of this method does not decrease as much as other methods while number of classes increased.