Research And Implementation Of Automatic Question Answering System

With the popularity of the computer and the rapid development of the Internet, online information is assuming the geometric growth, so we need to solve a problem urgently that how to retrieve useful information quickly and accurately in the mass of complex information. Traditional search engines have achieved some success, but the returning information of this engine is too cumbersome to meet the fast, accurate demand of people. Automatic question answering system applies some technologies such as network communications, artificial intelligence, information retrieval and natural language processing with intelligent, accurate and concise and can compensate for the shortcomings of traditional search engines effectively. Oriented the field of the national science and technology projects to declare, this paper focused on some key technologies of the automated question answering system. The main works are the following:(1)This paper puts forward a segmentation method based on the professional dictionary and ICTCLAS. At first, establishes the professional dictionary based on domain knowledge. Then splits the sentence using the positive maximum matching algorithm. Finally uses the tool of ICTCLAS dealing with all unknown words in the dictionary. Experimental result shows that the method has higher precision and recall rates, in particular the recognition of professional vocabulary.(2)The paper researches the similarity algorithm of words based on HowNet. First, calculates the righteousness original similarity of the concepts using the righteousness original distance of HowNet. And then makes the concept similarity of words. Eventually comes to semantic similarity of words.(3)A sentence similarity calculation method with multi-scale and various features is proposed. First, the paper improves the existing method of the TF-IDF approach based vector space model and the approach based semantic information. On this basis, taking Angle clause form, semantic and syntactic structure into account, considering six features of the sentence including word frequency, semantic, length, word form, word order and distance, the paper proposes a method to calculate the similarity of sentences with multi-scale and various features, and calculates the optimal portfolio weights using the genetic algorithms. Experimental results shows that the method has improved on the recall rates and precision rates compared with the existing sentence similarity calculation method.(4)The paper has designed and implemented a automatic question answering system to advisor the project application for the national science oriented library. The system has completed the establishment of the frequently asked question, the pretreatment of problem, the establishment of candidate problem sets and the computing of sentence similarity to verify the feasibility and effectiveness of the method that the paper proposed.