Research On Question-Answering System Mixed With FAQ, Ontology And Reasoning Technology

Internet technology has brought earth-shaking changes to human society, and people have got to get used to getting all kinds of information from the Internet. These are because of the development of the search engine technology. But there are still some unresolved problems to the search engine. Firstly, when the user searched the information, the search engine returned too much relevant and imprecise information. Secondly, the retrieval depended on the keyword matching, which could not actually express what the people really meant.Because of existing of these problems, question-answering system (Q&A system) emerges as the times require. Q&A system enables users to input a question in natural language and returns a succinct answer. Because Q&A system based on Web is so complicated that it's hard to focus on semantic understanding. Because of this reason, a domain-specific was selected and a strategy combined FAQ (Frequently Asked Question) method with Q&A method based on the ontology knowledge base was proposed in the dissertation. This dissertation emphatically aimed at the main technology of Q&A system such as question processing, answer extraction, semantic reasoning and some other technologies to conduct the research. The key content and innovation are as follows:1. To improve the performance of Q&A system, a model of multi-strategy question answering system was proposed. Combined with the advantage of question answering technology of FAQ base and ontology knowledge base, the question answering system of restricted domain based on this frame was realized. For the common questions, Strategy A was first used:the matching technology of user's input question and the questions of FAQ was used to realize mechanism of question & answer; for specific domain knowledge Strategy B was used:the domain ontology base was constructed, shallow semantic analysis and SPARQL query technology were realized. By definition rules of semantic chunk and decision rules of semantic chunk in shallow semantic analysis, the question vector was generated. Then, SQARQL query technology was used to search in ontology base. For a few questions that cannot be answered Strategy C was used:manual answering or relevant web pages were returned to answer the questions of users. The experiment result has shown that the question answering system based multi-strategy can improve the recall rate, precision rate and Fl measure-value of the system more efficiently than the system to use single-strategy.2. To retrieve knowledge of professional field, we created ontology of hospital domain and proposed a method of answer extraction of question answering system based on ontology base of hospital information, Firstly, the hospital information ontology base was constructed, then, shallow semantic parsing was used to parse the question, unknown and known information were indentified and the question vector was generated. After that, SPARQL query technology was used to search in ontology base and the answer was retured. Therefore, the recall rate of the returned answers got by the query of professional knoveledge was effectively improved and the performance of the system was reinforced.3. To improve the recall rate of question answering system and expand ontology knowledge base, a reasoning method of Jena reasoning used in city-domain hospital question answering system was proposed. In city-domain hospital question answering system, reasoning rules were firstly constructed and later added into Jena reasoning machine. Then, the results of the deduction were added into the knowledge base and as a result the knowledge base was expanded. This will help to get the answer which can not be searched in ontology base and can be got by inference of Jena reasoning, thus the recall rate of question answering system can be improved. In addition, to excavate some implied information of hospital ontology, a reasoning method used in question answering system to reason by the use of SWRL descriptive reasoning rules and Jess inference engine was proposed. The experiment has shown that the method can make computer have the ability of symptom reasoning, thus, the performance of question answering system is expanded.4. The pneposed model of question answering system was applied to hospital domain. The application helps to verify the relevant technology and algorithms of th(?)ubject. The experimental tests have shown that the system had higher precision rate and recall rate, therefore a valuable research was done to the design and application of question answering system in the thesis.