Abstract:
The problem of document classification has been widely studied in machine learning and data mining. In document classification, most of the popular algorithms are based on the bag-of-words representation. Due to the high dimensionality of the bag-of-words representation, significant research has been conducted to reduce the dimensionality via different approaches. One such approach is to learn a codebook by clustering the words. Most of the current word- clustering algorithms work by building a single codebook to encode the original dataset for classification purposes. However, this single codebook captures only a part of the information present in the data. This thesis presents two new methods and their variations to construct multiple non-redundant codebooks using multiple rounds of word clusterings in a sequential manner to improve the final classification accuracy. Results on benchmark data sets are presented to demonstrate that the proposed algorithms significantly outperform both the single codebook approach and multiple codebooks learned in a bagging-style approach.
