The Design of a Question Answering System for USENET

• Main Authors: Wei-Chuan Niu, 牛維娟
• Other Authors: Hsi-Chieh Lee
• Format: Others
• Language: zh-TW
• Published: 2003
• Online Access: http://ndltd.ncl.edu.tw/handle/74946375444029828639

碩士 === 元智大學 === 資訊管理研究所 === 91 === In recent years, web has become an important communication channel between people. Traditional search engine accepts a list of keywords inputted by user, and then replies a ranked list of search results. However, this way is limited by the keywords which user actually key in and, moreover, the lack of knowledge taxonomy information. Thus, user still needs to check the ranked list one by one to search the information he really aimed. The key point of the Q&A (Question Answering) system is about the ways to extract the information of a natural language question inputted by the user, and to find the correct category of text according to some knowledge taxonomy mechanisms and, finally, to extract the answer from those text. In this way, user does not need to pinpoint the category of the articles which he wants to search, and also does not need to know all the exact keywords related to the answer. USENET is a logical net composed of newsgroups all around the world. And there is rife with semi-structured Q&A messages. In this research, we investigated a Q&A system on IT domain newsgroups using an ontology-based knowledge taxonomy technique. We utilized WordNet synonymous set to extend the query from user’s input, and then find the interested category of the newsgroups by that information. Then we evaluated the messages in those newsgroups by a scoring mechanism. Finally, we extracted a referential answer from those messages with higher score. Besides, we designed different view functions to allow user to check more information related to the answer. Our result reveals that applying Q&A system to Usenet using a shallow natural language processing and word-overlap-based scoring function is applicable. But performance improvement still needs more knowledge resources.