Research And Implementation Of Question Classification Technology Based On Open Domain Question Answering System

As the rapid development of the Internet, People increasingly rely on search engines. Question Answering system, as the new generation of search engine, is an advanced form of the information retrieval system. It allows users to ask questions by means of natural language, and can supply more accurate answers. Comparing to the traditional search engine which is based on the keyword, question answering system can provide an accurate and concise answer to a natural language query. Question answering system generally consists of tree parts: Question Analysis, Information Retrieval, and Answer Extraction. This paper, base on open domain question answering system, focuses on the research of question classification, fact questions answer extraction and definition question answer extraction. Finally they are verified through experiments. The main works in this paper are as follows:Firstly, because of questions has a small vocabulary, a new classification method using Bayes and semantic chunk as category features is proposed in this paper. Semantic chunks are identified by question focus, question syntactic and the semantic roles of the question component. Besides, as the question focus chunk has important affect on question classification, the method combine with corresponding rules between question focus chunk and question type. Experiment result shows great efficiency.Secondly, this paper applies semantic role analysis techniques combines with question key-words on answer extraction towards factoid questions. And classifying question key-words into different sets, giving different weights according to their importance. Experiments show that the method improves the accuracy of the answer extraction on factoid questions.Thirdly, apply relations base on HowNet on answer extraction towards definition questions. Through analyzing the relation of each word in the candidate answer sentence and the word which needs to be explained. We can obtain the scores of candidate answer sentences, and extract the answer of the question. Feasibility is verified by experiments.