The Study of Applying Concept Map on FAQ Retrieval

• Main Authors: Hai-Sia Wang, 王海霞
• Other Authors: Sung-Shun Weng
• Format: Others
• Language: zh-TW
• Published: 2009
• Online Access: http://ndltd.ncl.edu.tw/handle/55170221509135470776

碩士 === 輔仁大學 === 資訊管理學系 === 97 === It is a conventional way using keywords on information retrieval. The disadvantages are information overloading, ambiguous query terms and imprecise queries. So far, domestic FAQ websites and the related literatures on FAQ retrieval mostly have used keywords and natural language for information retrieval. Because of the different expressions of users, it could have cognitive differences between users and systems so that users could not find any information. Therefore, using concept maps to represent FAQ questions with visualization for searching relevant questions will improve the disadvantages of using keywords and natural language. The purpose of this study is to apply the method of concept map for FAQ question retrieval. First, extracting the keywords to represent the questions by using k-means clustering algorithm for question clustering. Second, using association rules to produce concept rules in each question cluster. Finally, connecting concept rules to form a concept map. The measures of precision, recall and F-measure are used to evaluate the results of question clustering. The representativeness of concept maps is evaluated by computing the values of precision, recall and F-measure and also compared with the results of using keywords. The experiments of this study show that the results of using concept maps on information retrieval are not very significant, however, the performance on precision and recall is relatively higher than using keyword retrieval. Moreover, there is one problem with questions belonging to two clusters in this study in the collection of Taipei City Mayor Mail data and Chunghwa Telecom FAQ data. This problem is because artificially classified questions are not appropriate. This study proposes a new suggestion on classification. We explore the characteristics of two different data and the methods to be used in order to have better searching results.