According to the weakness of shallow semantics models based on lexical analysis which are commonly used in QA system, shallow semantics models cannot accurately analyze the deep semantics of users' questions. This paper focuses on the tourism QA application field, adopts the combined category grammar (CCG) to parse the question sentences, and uses lambda calculus to express the question semantics, so that semantic models on tourism questions can be derived. And it's convenient to search answers according to such accurate semantic quickly. The research first carries out data acquisition and corpus tagging preparatory work,including the analysis of tourism question corpus both in sentence pattern and syntax. Then the supervised learning process based on a probabilistic CCG algorithm is used to train a reliable semantic dictionary. At last, an automated answering system is built according to the semantic dictionary and related knowledge, which is mainly about the question parsing and building of corresponding semantic models. The final result on evaluation dataset shows that the semantic analysis method has relatively clear improvement in analytical performance.