Restricted-domain Chinese automatic question-answering system based on question sentence similarity

It is an available pattern to implement auto answer in RDAQAS (restricted-domain automatic question-answering system) through calculating the similarity of target question sentences and question sentences in question sentences corpus, and then finding the most similar question sentences, and retrieving the answer finally and all these are based on HowNet and domain ontology. This paper introduces the building of financial domain ontology and question sentences corpus, then proposes the method to calculate similarity of question sentences based on keyword vector space method and semantic concept vector space method. The procedure of realization is described in details. The learning algorithm and learning course of getting question sentence semantic vectors based on the maximum entropy model are also introduced in detail. Finally, the experimental comparing data illustrates that the similarity calculation method based on the semantic concept is more superior to that based on the keyword.