The Research And Implementation Of Question Answering System

Question answering system has become a hot research topic. Question answering systems always have question analysis, information retrieval and answer extraction, some question answering system based on FAQ have the FAQ module.This paper studies the two algorithms in question answering system, sentence similarity algorithm and answer extraction algorithm, designed and implemented a financial question answering system.Sentence similarity algorithm is a key algorithm of question answering system, not only for the FAQ module in question answering system can also be applied to the answer extraction module. This paper describes the existing sentence similarity algorithm and analyzes their advantages and disadvantages, then proposes an improved VSM algorithm, which used to calculate the sentence similarity between the user question and candidate questions. The improved VSM algorithm adds semantic information, and syntactic information in the traditional vector space model. The improved VSM algorithm doesn’t need to set weight for the characteristics of the sentence, reduce the calculation error caused by weight setting and improve the accuracy of sentence similarity calculation.In the answer extraction, the paper analyzed and researched the common answer extraction algorithms and proposed the answer extraction algorithm based on the window-based model. The user’s question filtered through two layers in this algorithm, the term characteristics of the sentences, semantic feature information and syntactic features added to the answer extraction process, and solved the problem caused by the segmentation system can’t recognize the phrase by the window-based modelFinally, the paper designed and implemented a financial automated answering system. The system consists of four modules, the question analysis module, the information retrieval module, the answer extraction module and the FAQ module. When the user submits a question, the system computes sentence similarity with the improved VSM algorithm, if the similarity is greater than a certain threshold, it will return directly the corresponding answer to the user. Conversely, system will do other operation for the user’s question; extract the correct answer for user. The question analysis module uses the existing rule-based combining syntactic analysis approach to analysis user’s question; the information retrieval module provides retrieval services by lucene. The system uses two ways to get the question’s answer, which improves the efficiency of the system; it can be automatically updated frequently asked question, which improves the accuracy of the system.