Applying Feature Opinion Pairs in Product Review Classification

• Main Authors: Po-shun Liao, 廖柏舜
• Other Authors: none
• Format: Others
• Language: zh-TW
• Published: 2010
• Online Access: http://ndltd.ncl.edu.tw/handle/94233843351678785540

碩士 === 雲林科技大學 === 資訊管理系碩士班 === 98 === With the rapid development of Internet, the rise of electronic commerce, and the extensive use of web2.0 technology, more and more people expresses their views of products and services on the Internet. However the Web''s comments are usually mixed with positive and negative comments, if processing manually to obtain information which is useful, it must spend a lot of energy and time. In text mining, document clustering can classify files to solve the current issue of disordering file on the Internet. When the use of text mining, the document will go through pre-processing of documents into a number of keywords to represent documents, and then converted into numerical values. General approach is to use vector space model, converted into vector according to each of the keywords, then use the clustering algorithm to cluster documents into categories. However, the use of vector space model, it is usually accompanied by excessive noise, making the effectiveness document clustering affected. In this study, feature representation method was improved by using feature-opinion pairs to represent the features of vector space model, so the feature representation method can contain more semantic information, and then generate the vector space model into classification to classify product reviews. The experiment results shows that the proposed feature representation method was successfully increase the five-way classification accuracy, in addition to the Average accuracy of setting root as Token feature for the extraction rule has a significant improvement, the accuracy in each star rating have a slightly increased. This study will be effectively add text polarity into the feature which makes the data distribution in feature space more clearly