Improve Natural Language Q&A System based on scenario Study

• Main Authors: Chen-Ming Wang, 王誠明
• Other Authors: Hsiu-hsen Yao
• Format: Others
• Language: zh-TW
• Published: 2003
• Online Access: http://ndltd.ncl.edu.tw/handle/75830561725258202131

碩士 === 元智大學 === 資訊工程學系 === 91 === People tend to find data or information through net surfing as computers and networks become more and more popular. Computer system analyzes user queries and searches for the demanded data or information. However, usually computer system provides only query input with fixed input column, while Q&A (Question and Answer) system merely uses the fixed-column query inputs as keywords to search and then to return the found data as results. This kind of Q&A system lacks of flexibility that user cannot express query condition adequately with limited input columns. To solve the problem discussed above, it is necessary that Q&A system have the ability to deal with natural language input, one language that is most familiar to users so that query conditions can be expressed exactly and adequately. Yet, computer analysis cannot handle the semantics ambiguity of natural language input without the aid of scenario, which will lead to misinterpretation. “Language communication is not just a language communication”. Human brain explains semantics ambiguity by the aid of scenario such as background knowledge, speaker’s body language, etc, which computer system does not have. Since scenario seems to have significant role in natural language analysis while there are only question strings provided by user in a Q&A system, we try to understand the scenario from lateral information of those question strings. For instance, length of string, keyword, use amount of proper noun, the relationship between subject of sentence and user, preceding sentences, etc… may be used to gather statistics and to analyze the scenario. According to the scenario, an appropriate natural language logic handling is applied while parsing and comparing the natural language. Proper update to keyword database, weight of keyword, syntactical and morphologic features similarity comparison algorithm, is also performed then. This mechanism approaches the logic of human brain in the way that, not only one natural language analysis rule is applied stiffly to any user, but, an appropriate analysis rule is selected based on natural language query and its corresponding scenario and then is applied to distinct users, and thus may reduce analysis error.