Design And Implementation Of Automatic Question Answering System For College Entrance Examination Based On Knowledge Graph

In recent years,the number of college entrance examination candidates has reached a new high.The college entrance examination is the most important opportunity for young people in our country to change their destiny.After the college entrance examination,it is very important to fill in the aspiration forms.College entrance information is one of the main sources of information for candidates and parents to choose colleges and majors after the college entrance examination.If candidates can obtain timely and accurate college information,they will be more likely to enter their ideal college and major.Therefore,this paper focuses on the research on the automatic question and answer system for college entrance examination based on the knowledge graph,which mainly includes the following aspects :This paper discusses the architecture of the automatic Q&A system for college entrance examination,including knowledge graph construction module,and Q&A query module.This paper designs and implements a Q&A system for college entrance based on the knowledge graph.In this paper,we defines 5 entities,4 relationships and 11 attributes of the knowledge graph.This paper discusses the algorithm of the Q&A system,which is divided into entity recognition tasks and question and answer matching tasks.The traditional word-embedding model in the task of named entity recognition may inadequate extraction of semantic information of words.Aiming at solving shortcomings of traditional model,we use the pre-training language model ALBERT to replace the traditional word2 vec in training word vector.Also,only using character vectors or word vectors in the entity recognition task can make our model unable to fully extract deep-semantic features.So The CAW algorithm of fusion word vector is introduced.CNN is used to train word vector in order to obtain word semantics and Bidirectional GRU is used to initially extract contextual information.Then we splicie both of them and sent to the Bidirectional LSTM to locate the entity position in the question,and finally the entity recognition result is output through the CRF layer.The experimental results on the People’s Daily and the cluener dataset show that the use of this model can effectively improve the accuracy of entity recognition.In the task of question and answer matching,we use the pre-trained language model ALBERT and the Bidirectional GRU to train text to obtain the semantic vector representation of the question.The distributed vector representation model TransH for the knowledge graph cannot handle the one-to-many or many-to-one entity relationship types,so we use Bi GRU and the distributed vector representation model Tran H of the knowledge graph on the hyperplane to obtain the semantic vector of the knowledge entity and relationship,and finally select the most correct knowledge triplet by calculating the cosine similarity of the question,entity and relationship.Experimental results show that the use of the question answering matching algorithm in this paper can improve the accuracy of question answering.This article conducts experiments on the basis of the above to prove the reliability of the algorithm model,collects information of more than 3000 colleges and universities,and realizes the automatic question and answer system for college entrance examination based on the knowledge graph,and the actual operations show the effect.The results show that the automatic Q&A system for college entrance examination based on the knowledge graph designed and implemented in this paper can meet the needs of users for obtaining professional information in colleges and universities,and has good feasibility and practical value.