Two Stages Text Classification via Smooth Support Vector Machine

• Main Authors: Chih-Hung Tsai, 蔡至泓
• Other Authors: Yuh-Jye Lee
• Format: Others
• Language: zh-TW
• Published: 2005
• Online Access: http://ndltd.ncl.edu.tw/handle/72724320427534565170

碩士 === 國立臺灣科技大學 === 資訊工程系 === 93 === In recent years, text classification is one of the popular research topics. The technique can be extensively applied to many cases such as information retrieval, email filtering, topic-based retrieval, and personalization of search engines, etc. In this thesis, we propose a two stages linear classifier framework using smooth support vector machine (SSVM) for text classification. This framework contains two classifiers. The first classifier is generated by SSVM and b-tuning procedure. Its main purpose is to increase the recall. The second classifier tries to enhance the performance of the first classifier. We also try several different feature selection methods, mutual information (MI), χ2-test and DF-Entropy, to select terms as features, and to compare their classification results. It turns out that the DF-Entropy method has the best performance among these methods. Besides, we also utilize linear reduced support vector machine (RSVM) for text classification. This approach is to use linear kernel to replace non-linear kernel in RSVM. Its advantages are that it needs not do feature selection and can reduce computational time. From the comparison results, two stages linear classifier has the best performance and linear RSVM also has good performance.