Question Answering System Based On Web Search

With the rapid development of Internet, the network has become animportant source of information. How fast and easy to get information fromnetwork has become the focus of research. Although the search engines andautomatic question answering system has solved this problem, there are alsosome shortages of them. Searching information with keywords, search enginecannot express the intention of users clearly. And the result of search engine is aset of relevant pages, so it needs users to find the final answer by themselves.However the traditional automatic question answering system can improve theseshortages of search engine, it can accept the questions inputted by users, andreturn the best answers. But the traditional automatic question answering systemneeds to keep a large knowledge base, so the information coverage and updaterate became the main shortage. To make up this shortage, we proposed a questionanswering system based on network environment, which uses the internet as theknowledge base, and search information with search engine. Finally, according tothe input of users, this system will extract the correct answers and return them tousers.According to the structure of traditional question answering system, thispaper has designed the overall framework of the system, and allocated tasks foreach module. In question analysis module, the main works are questionclassification and keyword extraction. And in information retrieval module, weused search engine to retrieve related web pages, and downloaded them. Besides,according to the type of question and source of information, we putted forwardsome different policy for answer extraction and scoring. Furthermore, in thispaper, the named entity recognition technology has been used to extract answersfor fact kind of questions, and the answer similarity calculation method has beenused for non-fact kind of questions. Finally we calculated scores for eachanswers, and selected the one with highest score for users.In addition to the above work, we also made some improvements. First theoriginal question classification model has been improved for several special questions. Then we also extracted qualifiers from questions with syntacticdependencies, and set different weights for each keyword. Finally, with thedifferent weight of keywords, we improved the answer scoring method and theanswer similarity calculation method.Base on the building and implementation of this question answering system,this paper used the artificial building questions set to test this system’sperformance, and evaluated the improvements of the question classification andanswer extraction. The finally experimental data has shown the effectiveness ofthese improvements. And the effect of the actual operation also proved thefeasibility of this system.