Study On Chinese Question Answering System In Travel Domain

Because of the rapid development of information technology, it becomes more and more quickly and easily to store, exchange and search information. However, the huge amount of data and its disorderly growth has also brought new challenges to the traditional search technology. In such conditions, question Answering (QA) system becomes a hot topic in research and applications. The research of QA system is a field that embodies the intelligence of computer. Different from traditional information retrieval systems such as searching engine, it uses accurate, concise and natural language to answer user's questions which are also natural language. QA system is a comprehensive use of natural language processing technology. Generally, the main part of a QA system includes question processing, information retrieval and answer extraction. From the respect of answer extraction, QA system can be classified as frequently asked question based system and answer extraction technology based system.In this paper, referred to the philosophy of frequently asked question based system, we developed a structured matching and semi-structured query based automatic QA system in travel domain. System's core algorithm is sentence similarity calculation which plays an important role both in question matching module and answer extraction module. We introduced and compared some sentence similarity calculation methods and proposed our own method which is suitable for travel system. With the use of HowNet semantic dictionary some statistical information extracted from corpus, we design the method based on fuzzy set. On the other hand, we combined question template with free text searching to treat with non-frequently asked questions. Sentence similarity calculation method is also used in text search technology.With a friendly interactive interface, we implemented the system functions. From the perspective of administrator, the system provided question and answer database management functions. From the perspective of user, the system is divided into direct search of frequently asked questions, template matching based search, and non-template search of other questions. The system also provided a user feedback mechanism in order to improve and update our database. At last we tested system's function and evaluated experimental results for different types of questions.