Novel Approaches to Pre-processing Documentbase in Text Classification

Abstract. Text classification is currently popular in Knowledge Discovery in Databases (KDD) and Machine Learning (ML). KDD based text classification research focuses on statistical techniques, while the ML based approach focuses on artificial-intelligence techniques. Text mining necessitates the pre-processing of the documentbase. Two broad approaches can be identified: (1) document representation and (2) feature selection. In this paper, we propose three text pre-processing approaches: (a) pruned (two-frequency-threshold based) bag of single-words; (b) emerging pattern based bag of single-words; and (c) bag of frequent-wordsets. The core strategy to develop these three approaches is to avoid computational overhead of (2). Experiments indicate acceptable text classification accuracy when applying these approaches.