Research On Knowledge Graph Question Answering System Based On Contrastive Learning

In this era of rapid expansion of information,the Internet has become a huge repository of information.It is quite difficult to find the desired information among numerous data,and search engines are often used to assist in this process.However,search engines that use keywords to search will return a large number of related information links,including a lot of useless information and even advertisements,making users waste a lot of time and effort in filtering information.Knowledge Base Question Answering(KBQA)is one of the improvements to KBQA.KBQA uses Natural Language Processing(NLP)technology to analyze the semantic information of user query sentences,understand the main body and intention of user queries,and then accurately return answers through knowledge maps instead of a large number of information links,greatly improving user retrieval efficiency.Currently,the complex and variable forms of user questions,semantic ambiguity,and the existence of entity aliases make it difficult for computers to accurately locate the corresponding entities in the knowledge base and understand the user’s consultation intentions.Based on Comparative Learning(CL)to optimize the semantic vector representation of user sentences,this paper analyzes and studies open domain knowledge map question answering systems and restricted domain knowledge map question answering systems.The main work is as follows:(1)In the field of open domain KBQA,there are abbreviations,aliases,and nesting of entities in question sentences,as well as the gap between the structured semantics of question sentences and knowledge bases.Most studies regard entity disambiguation and relationship matching as error transmission caused by independent subtasks,resulting in poor overall system accuracy.This paper proposes a semantic union model(SUM)based on the BERT model combined with comparative learning,which combines entity disambiguation and relationship matching in a unified framework.It can provide information for entity disambiguation through entity affiliation,and simultaneously complete entity disambiguation and relationship matching tasks,avoiding error transmission,and improving the overall performance of the system.It can provide information for entity disambiguation through entity affiliation,and simultaneously complete entity disambiguation and relationship matching tasks,avoiding error transmission,and improving the overall performance of the system.The results of simulation experiments show that the semantic joint model based on comparative learning has some improvement in the field of open domain knowledge map question answering systems.(2)This paper uses medical disease knowledge to construct KBQA,and in view of the difficulties in obtaining data,uneven intention categories and noise in the limited domain KBQA domain,which lead to the problem that the model can not accurately identify the user’s intentions.The traditional classification task of intent recognition is transformed into a semantic similarity calculation task,which is fine-tuned using a comparative learning model.By comparing the loss,a more distinguished semantic vector representation is obtained.Further,it can more accurately identify user intent and improve the accuracy of knowledge map question and answer system.The results of simulation experiments show that the model based on comparative learning has some improvement in the field of limited domain knowledge map question answering systems.