Research On Open-domain Chinese Question Answering System Based On Web

Since the 1950s when artificial intelligence initially was researched, the question answering systems drew much attention of the researchers. At present, applied English open-domain question answering systems have emerged. On the contrary,the researches on Chinese open-domain question answering systems are relatively laggard. Hence in this paper two schemes of Chinese open-domain question answering systems based on Web are presented,and the feasibility of them are verified throught experiments.In the first scheme, firstly a new question classification architecture based on HowNet is established .In the stage of question analyse, according to the type of the question query-generation module generates one query which can be used by Google. In the stage of information retrieval, Google uses the query to get the correlative documents,then extracts some documents as the candidate documents. In the stage of answer extraction, the candidate answers choosing module combines HowNet and Named Entity Identification technique to extract the candidate answers from the candidate documents. Finally every candidate answer will be verified through the query results of the Google and the candidate answer which has the highest reliability will be the returned as the best answer.In the second scheme which doesn't need to establish the question classification architecture in advance, after analyzing the type and function of the part-of-speech of the question the unique sentences bank of the Chinese FrameNet will be used to determine the semantic type of the answer and contract some queries. In the stage of information retrieval,Google uses the queries to get the correlative documents and extracts some documents randomly as the candidate documents. In the stage of answer extraction,the candidate answers choosing module extracts the candidate answers through analyzing the correlative documents and verifies every candidate answer throught the query results of the Google. Finally the candidate answer which has the highest reliability will be returned as the best answer.The main discrimination between the two schemes is the method of the query construction.The speciality in this paper is that a new classification architecture is presented which can be implemented in the condition of the current techniques.