Research And Implementation Of Intelligent Question Answering System For Chinese Discourse Question Based On Deep Learning

With the wide application of deep learning technology in the field of Natural Language Processing,the research focus of intelligent question answering system has changed from traditional document retrieval in question library to problem matching and answer reasoning based on natural language question understanding.Firstly,we analyze and research the key technology used in this paper,and then according to the structural characteristics of discourse problem,we proposed a hierarchical semantic model based on the CNN and Bi-LSTM deep learning model.This model integrates the semantic understanding of factual part into the understanding of the core question and finally get the integrated semantic understanding vectors which have more obvious features and less redundant information and have higher similarity in vector space.Based on the semantic understanding vector of this model,this paper proposed cluster storage model based on the improved algorithm of K-means,and two-way feedback answer recommendation algorithm based on massive data similarity calculation used Annoy toolkit which solves the data storage,the answer recommendation,effective accumulation of data and other core applications in intelligent question system.Finally,combined with the above model algorithm,we propose a design and implementation of high availability cluster architecture for intelligent question answering system,which includes application system cluster,Dubbo distributed service cluster,Hadoop computing cluster and database storage cluster and complete the system test.Experimental results show that the hierarchical semantic extraction model in this paper has better performance in accuracy and vector similarity than the traditional semantic understanding model,and the design of the application of intelligent answering system cluster architecture can support the need of massive data scene and has practical application value.