Research On Open-Domain Question Answering System

Open-domain question answering has emerged as a new field aimed at the extraction of precise answers from large text collections in response to natural language questions.This dissertation systemically addresses the different components of Question Answering System. Question Analyzing Component makes use of heuristic rules or machine learning algorithms to identify the right type of information that the question requires. The role of Information Retrieval Component is to narrow the relevant information. Since there exists too much redundant information in relevant documents, sentence retrieval algorithms are incorporated to find the most relevant sentences. Answer Finding Component extracts candidate answers with the Information Extraction techniques, and then selects a most possible answer from them.The Question Answering System ICTQA we built has proved to be fairly acceptable with the performance in QA Track of TREC 2003 and we describe its architecture and function modules in details. On the basis of the system performance analysis, we take fully account of various document retrieval systems, sentence retrieval algorithms and retrieval granularities respectively and make an improvement of 45.45% in the Information Retrieval Component. Furthermore, we take complicated questions into account and use delicate strategies to solve specific types of complex question.