A Feature Selection Method Based on Text Segmentation of E-Books

• Main Authors: Ming-WeiLai, 賴銘偉
• Other Authors: Hei-Chia Wang
• Format: Others
• Language: zh-TW
• Published: 2010
• Online Access: http://ndltd.ncl.edu.tw/handle/02023913522965767774

碩士 === 國立成功大學 === 資訊管理研究所 === 98 === With the exponential growth of information technology and Internet, paper books can be transformed into e-books. People can get these e-books form Internet and download them by e-reader. It enhances the convenience to absorb knowledge from books. However, the number of e-books has been very large. It costs lot time and energy to classify these e-books. The traditional classification approaches like decision tree, k-nearest neighbor, na?ve bayes, support vector machines, usually select the feature words from content. These words will form a feature space. The longer the article is, the more likely generate a lot of feature words, and the dimension of feature space is higher. It causes the follow-up of the classification process complicated. Therefore, the classification process steps filter undesirable feature words through feature selection. However, the length of e-books is usually much longer than the general article. With traditional approaches, e-books generate a large number of feature words, and cause the follow-up of the classification process complicated, and even lost important feature words because of the long length content reducing these words’ overall weights. Therefore, we present a novel feature selection approach which applies a text segmentation algorithm. With this algorithm, e-books can be cut several segments.We analyze all words’ importance in these segments and select the inportant feature words for every segments. We expect that the feature words which selected by our approach can improve the accuracy of classification.