Research On Evidence Retrieval And Scoring Algorithm Based On Chinese Deep Intelligent Question Answering System

Accompanied by in-depth research in technologies such as knowledge management and intelligent depth analysis,the deep intelligent question answering(QA)system based on knowledge level have gradually become an indispensable part of the development of artificial intelligence.The deep intelligent QA system,which mainly adopts the combination of Deep QA framework structure and knowledge map reasoning,uses cascading cooperative processes to handle problems and has greatly improved the system framework extension and intelligent reasoning analysis..The advantages and disadvantages of evidence retrieval and scoring algorithms play a key role in the accuracy of the system.After in-depth research and analysis,this thesis clarifies that there are some problems that need to be solved in the evidence retrieval and scoring algorithms based on the Chinese deep intelligent QA system.Firstly,the current Chinese deep QA system lacks an evidence scoring algorithm based on syntactic structure and semantic analysis.Secondly,there is a lack of efficient evidence parsing strategy and entry preprocessing process in the existing algorithm flow.Thirdly,the existing evidence retrieval and scoring module lacks a better merge algorithm based on a paragraph scoring pool.Therefore,optimizing and innovating the evidence retrieval and scoring algorithm is one of the important issues to improve the performance of the Chinese deep intelligent QA system.In order to solve various defects and deficiencies in the existing evidence retrieval and scoring algorithms,this thesis proposes a novel evidence retrieval and scoring algorithm system that is applicable to the Chinese deep intelligent QA system.First of all,the method of generating the evidence passage based on the free document and the preprocessing strategy of the lemma are added.Then,two kinds of evidence scoring algorithms based on term frequency and lexical order were improved,and two new algorithms,which are syntactic structure scoring algorithm based on convolution tree kernel function and semantic analysis scoring algorithm based on language representation model and convolutional neural network,are proposed.Finally,based on the study of the merge algorithm of the current paragraph scoring pool,two new scoring pool merging algorithms based on PCA and K-means are proposed.Experiments show that the improved evidence retrieval and scoring algorithm improves the system algorithm system and reduces the cascade process error,effectively improving the accuracy of the entire system.