Research On Intelligent Question Answering System Based On Deep Learning

With the era of big data, people are faced with information explosion and information overload. Search engines which return a large amount of information has been unable to meet people's need to get information fast and accurately. The emergence of intelligent question answering system make up for the lack of search engines to a large extent. Therefore, some people even assert that question answering system will be the next generation of search engine. Open domain question answering system,specific domain question answering system and chat robots, which have greatly improved the efficiency of people's access to information and reduced the labor costs of enterprises.However, the existing intelligent question answering systems are not sufficient for the application of the deep learning algorithm. Most intelligent question answering systems of commercial companies are relatively closed, not open-source. In this paper, we studied the existing question answering system and the deep learning theory in depth. We hoped that the newest algorithm of deep learning theory could be used in research of intelligent question answering system in order to improve the accuracy of question answering system.The main research of this paper includes the following aspects:First: In this paper, a new deep learning algorithm which is called Deep Boltzmann Machines is applied to the construction of knowledge base. We not only analyzed the structure of Deep Boltzmann Machines in detail, but also proposed a knowledge base construction model based on Deep Boltzmann Machines. Compared with the artificial neural network(ANN) algorithm and the deep belief network(DBN), the experimental results showed that the model had a better experimental effect, and improved the accuracy by about 4%.Second: In the information retrieval module, this paper proposed a deep-mixing algorithm (SM-BLSTM) based on semantic model and stacking bidirectional long short-term memory model. This model which utilizes semantic features of semantic model and the depth structure with cycle structure of bidirectional long short-term memory, can fully expresses the temporal characteristics of sentences. At the same time,compared with maximum entropy model, artificial neural network and convolution neural network algorithm, SM-BLSTM achieves better experimental results in different data sets, the average accuracy rate is improved by about 5%.Third: After the modeling and analysis of the intelligent question answering system was completed, this paper presented an intelligent question answering system for the field of film knowledge based on the above model. Compared with the existing intelligent question answering system, our system achieved better experimental results, and could improve F1 measure by 4.3%.