The Fact That Class To Answer

This paper presents research and implementation of Factoid Question Answering System. A Factoid Question Answering System involves: PreProcess, Question Analysis, Document Retrieval, Answer Extraction, Answer Ranking and Answer Projection. In this paper, the module organization of the system is discussed in detail, implementation of these module as well.This paper looks deeply into the Document Retrieval and Answer Ranking Module especially and proposes effective methods to improve the performance of these two modules.The Document Retrieval module acts as a basis of all the posterior processing of the system. We discuss the importance of the query expansion method in document retrieval module. Then we present a query expansion implementation based on automatic relevance feedback for the document retrieval module. Steps in query expansion implementation are discussed in detail:: fetching initial relevant document; extracting expansion terms form initial document and formulate the extended query. When calculating the confidence of the candidate expansion term, we introduce Wordnet as knowledge base to do some adjustment to the confidence score. The experiments show that query expansion improved the quality and redundancy of the relevant documents returned by the Document Retrieval module .Another important module is Answer Ranking module. The effect of Answer Ranking module determines the performance of the whole answering system. We present a answer ranking model using syntactic analysis and statistical learning.. The evaluation function in the ranking model is trained by using the SVM model and leverages two kinds of feature: density-based features and feature using syntactic analytic. The experiments show that the evaluation function can give good confidence score to candidate answer and improves the performance of the Answer Ranking module.