Research On Key Technologies Of Intelligent Question Answering Based On Domain Knowledge Graph

In order to adapt to the increasing application of network information retrieval,the research on algorithm intelligent question answering based on knowledge graph is becoming more and more active.The knowledge graph plays an important role in intelligent services,but the complex semantic relationship brings great challenges to knowledge extraction,so the accuracy of question answering needs to be improved.Therefore,this paper takes knowledge extraction from knowledge graph as the research topic,focusing on the realization technology of relation extraction model in different scenarios and the influence of relation extraction on the accuracy of question answering.The main work and contributions of this paper are as follows:1.A hierarchical relation extraction model based on Bi-LSTM is proposed.The traditional model usually solves the general classification problem,which results in the lower accuracy of the extraction.In this paper,hierarchical semantic information between relation labels is used for modeling,and an extraction model based on hierarchical loss function is proposed to solve the shortcomings of semantic features in the model back propagation.Experiments show that this method can effectively improve the performance of knowledge extraction compared with CNN,RNN,and other traditional extraction models.2.A few-shot relation extraction model based on context attention mechanism is proposed.The current model is mostly based on supervised learning and distant supervision,which cannot be applied to use the scenarios when data is insufficient.This paper uses the method of few-shot learning and the context attention mechanism to construct the model.The experimental results show that the proposed method can carry out relation extraction even when data is sparse.3.A question answering model based on knowledge embedding attention is proposed.Relation prediction is one of the subtasks in question answering tasks using a knowledge graph.The existing model does not make effective use of candidate entities and relations,making the accuracy of question answering not high.This paper proposes to use knowledge embedding attention to fuse candidate entities,candidate relations,and question texts to improve the accuracy of relation prediction.Experimental results show that this method is superior to the existing question answering model.4.An intelligent question answering system based on a medical domain knowledge graph is designed and constructed.Through the custom medical entity and relation type,each module of the system is built,and the initial scale of medical domain knowledge graph and intelligent question answering system is formed.