Algorithm Research Of Automatic Question Answering Based On Semantic Comprehension

With the rapid development of the Internet,the information on the Internet has become more and more abundant.It is also increasingly difficult to find the information needed quickly and accurately in these vast amounts of information.Although with the use of search engines,people can easily get all the information they want.However,with the increasing number of online information and increasingly complex search expressions,users find it hard to search the required information from a large amount of content returned by search engines,so in order to meet people's higher retrieval needs,automatic question answering system gradually developed.The main work of this thesis is as follows:By analyzing the existing problems in the field of automatic question-answering,this paper proposes further ideas and innovations in the field of semantic understanding and semantic relevancy.In combination with the recent rapid development of deep learning techniques,the neural network is used to construct the model of this thesis.First,based on the Deep Structure Semantic Model,we improve our deep match model,this results in a hash index with minimal loss of accuracy.And on the rank model,the Compare-Aggregate neural network architecture is selected,and based on which,we propose the improvement innovation structure.The Dynamic-Clip Attention structure is proposed to replace the traditional attention mechanism,and the ranking result is improved by filtering the noise in the semantic association process.In the course of optimizing training,Learning to rank is used for reference,and Listwise is used to enhance the robustness and fitting ability of the model.Therefore,this thesis is the algorithm research about automatic question answering.The main purpose is to implement the semantic realization between question and answer.By improving innovation,we achieve more efficient match algorithm,and through the network structure optimization,we can provide a more accurate match,and we also validate the experiments on multiple data sets.This study helps us to build a highly efficient and accurate automated question answering system for practical scenarios.