| Compared to traditional search engines, Automatic Question Answering System is theoretically better to meet the needs of the user's search. However, questions from the real-world for the Question Answering System are often complex, which is mainly due to the lack of abundant world knowledge as well as powerful reasoning technology on natural language.As a modeling tool which can describe the conceptual structure of information system at semantic and knowledge levels, ontology has very large advantage in knowledge modeling. Taking advantage of ontology is bound to raise the performance of Automatic Question Answering System.This thesis aims to research how to use the semantic information of domain ontology to improve the performance of Question Answering System. On the one hand, the performance of the knowledge-based Automatic Question Answering System depends largely on the merits of the knowledge base, and therefore this thesis expounds how to create a well-structured domain ontology knowledge base for Automatic Question Answering System. On the other hand, the Automatic Question Answering System based on ontology knowledge base needs necessarily some corresponding technologies, and therefore this thesis proposes the method of question-analysis for domain ontology as well as the approaches of querying and reasoning on domain ontology. Finally, experiments are conducted to confirm the feasibility of the technical routes.(1) This thesis chooses the field of tour for research, creates a more rational domain ontology framework for Automatic Question Answering System, and selects a number of famous sights in Shanxi Province as instance data of the ontology. At the same time, the thesis makes exploratory study on the method, procedure and principles of creating domain ontology.(2) Because traditional classifications for open domain questions are not suitable for a specific domain, the thesis proposes a classification for domain questions. And then a method of extracting structured semantic information of questions based on CFN labeling is proposed. In the process of answers' extraction, the thesis proposes the method of questions'semantic information mapping to the ontology and the way of using inference properly.(3) In order to confirm the feasibility of the series of methods proposed, the thesis establishes an experimental Automatic Question Answering System. And then the experimental results show that the proposed methods are effective. |
