Research On Intelligent Question Answering Method Based On BERT And Knowledge Graph Embedding

Intelligent question answering aims to directly let the machine understand the users natural language questions and answer accurate answers.In recent years,knowledge graph and deep learning technologies have developed rapidly.Deep learning can understand user semantic information more effectively,and structured knowledge in knowledge graphs can provide users with more accurate information.For simple questions that usually only contain a single entity and relationship,Aiming at match natural language problems with the structured knowledge in the knowledge graph,this paper designs an end-to-end question and answer method based on the pre-trained language model BERT and the knowledge graph embedding,and completes an intelligent question answering system that enables users to acquire knowledge accurately.The main work of this paper is as follows:1.This paper proposes a head entity detection model based on pre-trained language model.Aiming at the key problem of knowledge graph question answering,question entity detection,this paper introduces the pre training language model Bert and its variants,and improves its output,so as to better extract the text features of natural language questions and get their embedded representation,which effectively improves the problem of polysemous words and out of vocabulary.At the same time,bigru network and attention mechanism are added to the model structure to obtain deeper text features,which effectively improves the accuracy of head entity detection.The model has achieved 98.67% and 98.91% F1 values on the Simple Questions and NLPCC datasets.2.The structured knowledge in knowledge graph is integrated into the deep neural network model.Aiming at the problem of multiple expressions of relations in different questions,this paper proposes an end-to-end knowledge graph question answering method based on knowledge graph embedding.Unlike the traditional word embedding,which only preserves the semantic representation of words,knowledge structure information is stored in knowledge graph embedding.Through the deep neural network training,the knowledge graph embedding representation and the natural language question embedding representation are linked together,and the answer selection is completed through a well-designed joint distance formula.The advanced performance of the model is verified on the Simple Questions and NLPCC datasets.3.Based on the knowledge graph question answering method proposed in this paper,an intelligent question answering system is realized.The system has a good front-end and backend design,which can answer user questions in real time and display the results visually,improve the efficiency of users accurately acquiring knowledge...