Answer Selection Method Of Question Answering System Based On Deep Neural Network

With the rapid development of Internet technology,people have more and more channels to obtain information,and these fragmented information is easy to be forgotten.In addition,due to the popularization of digitization,people's demand for knowledge inquiry is also increasing.Through the Q&A system,users do not need to retrieve answers from the long text or product catalog as traditional search engines do,but directly get more precise,short answers.With the development of computer technology and the application of deep learning in natural language processing,the research of question answering system has a new direction.The language model based on neural network can express the question and corpus as semantic vectors,so that the matching answers can be found automatically from the corpus.Compared with the previous,the deep neural network structure effectively solves the tedious system design and repeated data collation work of building the expert system.In this study,a hybrid question answering system based on information retrieval and deep answer matching is implemented based on the review of current research on question answering system.The information retrieval algorithm is used to select the partial answers with key information from the large-scale answer database.Then the answer matching algorithm model based on deep neural network is used to calculate the similarity between some answers after preliminary screening and users' questions,so as to get the best answer.The work of this paper is as follows:Firstly,a answer selection model based on Bert is constructed.In this paper,the TREC QA dataset and Wiki QA dataset are used to integrate the Bert algorithm and the idea of transfer learning.Bert is used to realize answer selection,and the experimental results of Bert on TREC QA dataset and Wiki QA dataset are evaluated.Secondly,an answer selection model based on MESIM is constructed.MESIM algorithm integrates the respective advantages of Transformer algorithm and ESIM algorithm,and achieves a good index in the final result evaluation.Experimental results show that although BERT algorithm has achieved good results in the index,its performance consumption is huge,response time is slow,and user experience is poor.The MESIM algorithm proposed in this paper achieves the same index as Bert,but consumes very little computing resources and has a fast response time,so it is an effective algorithm model for question answering system.Finally,a hybrid question answering system based on information retrieval and reordering is constructed according to different deep neural network algorithms.The question answering system constructed in this paper uses information retrieval technology to screen users' questions,and then uses the answer selection algorithm based on deep learning to carry out semantic matching.Achieve the framework of the whole Q&A system,as well as the evaluation of the whole system.