Question Analysis In Automatic Question And Answer System

With the development of the internet techniques, the number of online information is increasing with an explosive speed. Considering current Single Search Engine has its limits, we have been developed as a flexible open-domain QA framework which allows efficiently implement and combine new techniques for question analysis based on multi-SE (Search Engine). The main goal is to retrieve explicit answers to questions rather whole documents. In this paper we give an overview of the system.Instead of focusing on the optimization of a single approach, we combine several techniques for question analysis to fit the requirement of users. Individual techniques usually have weaknesses regarding their precision and the types of questions they cover, thus we integrate multiple approaches to build a strong overall system. For the factoid and list questions, we propose a simple approach based on answer type analysis with a more sophisticated pattern learning approach together. This first approach determines the expected answer type of q question from a hierarchy of Named Entity (NE) types and chooses an appropriate NE tagger to extract entities of that type. While this approach has a high precision, it fails if the answer type cannot be determined or the answer cannot be tagged. Thus, we complement it with another approach that uses textual patterns to classify and interpret questions. The interpretation of a question from its original formulation but yet preserves its semantics. Meanwhile, we combined the two approaches with a number of backup techniques for query formulation that are used when the first two approaches fail.In this paper, we implement the experiments on the system through selecting questions from different domain, questions with or without interrogative words. We also analyze the system's performance in QA track. The Precision of the system is 53% and the Mean Reciprocal Rank is 36%. The results are encouraging and showing that system has a high precision and can satisfies the basic requirement of users.