Performance Optimization Study Of Intelligent Question Answering System Based On Deep Neural Network

The intelligent question answering system is an artificial intelligence information service system that integrates natural language processing,information retrieval,and semantic analysis.This system,in the form of a question with an answer,can generate reliable and accurate responses automatically by processing and analyzing the input sentences.It provides the service as close as possible to the characteristics of daily interaction between people and gives personalized responses for each user.Intelligent question answering systems give computers the ability to understand and respond to human sentences.Besides,it has the advantages of wide application range and being easy to use.However,the expression of dialogue is flexible and diverse,at the same time,dialogue is constantly being updated through people's production and life.So for computers,there are huge technical challenges and difficulties in using traditional methods to directly analyze human sentences and respond.Most of the existing intelligent question answering systems can only implement simple functions and can not play a role in real scenarios.In addition,there are still some defects such as unanswered questions and limited response scenarios in the response process.In recent years,with the rapid development of artificial intelligence technology and deep learning,various ingeniously designed neural networks have emerged,such as convolutional neural network,recurrent neural network,deep neural network and so on.These technologies help researchers find new breakthroughs in the field of natural language processing.In order to solve the above problems and create a system with practical value and smooth response,in this paper,a design method of intelligent question answering system based on deep neural network is proposed and optimized.The existing intelligent question answering system mostly uses online open source corpus such as Google corpus,then researchers perform a series of preprocessing on the corpus,including text cleaning,word segmentation,part-of-speech tagging,word vector representation and word vector weighting.After these processes,the system generates a response by searching for matches.We optimize the existing intelligent question answering system from four aspects.First,in terms of corpus establishment,in addition to using daily chat data published online,we collected real data on commercial promotion themes and front-end design themes in real life scenarios.This corpus can ensure that the intelligent question answering system can provide value to real life.Secondly,in the preprocessing process,instead of using the bag-of-words model to represent the word vector alone,we combine it with the Skip-gram model to generate the final word vector together.The word vectors generated by using bag-of-words model alone are sparse vectors and independent of each other.Although this representation is friendly to discrete features,the encoded word vector loses the connection between words.The Skip-gram model uses the Euclidean distance between words in vector space to indicate how similar the two words are.This method increases the amount of information contained in each word vector,facilitates the machine to understand continuous sentences,and generates dense vectors to avoid dimensional disaster.Using the hybrid word vector expression model can make up for the shortcomings of using bag-of-words model alone,while retaining the excellent discrete feature processing capabilities of bag-of-words model.In the third aspect,the system optimizes the word vector weighting processing.Discarding the traditional method of determining the weight of word vectors based on word frequency alone,we use the term frequency-inverse document frequency weighting method to improve the weight of key words,and then output the final word expression.At last,the answer is produced using hybrid retrieval model and generate model based on attention mechanism.Compared with using the retrieval model alone,the hybrid model can not only answer the predefined questions in the corpus,but also truly understand the meaning of the sentence through the training and prediction of long short-term memory network.The hybrid model can supplement the question of answering the open domain.Experimental results show that the hybrid word vector expression model and the word weight inversion method using the term frequency-inverse document frequency can improve the quality of corpus preprocessing.The response generated by the hybrid model can meet daily practical needs and the entire system can communicate with people smoothly.