Research On Intelligent Question Answering System Based On Knowledge Graph Of Basic Education

With the rapid development of the Internet,the era of big data has arrived.The traditional search system based on keyword combination cannot parse the semantic information of user questions,but simply returns relevant answers for user reference.Thus the intelligent question answering system based on knowledge graph has emerged,which can get the answers quickly and accurately according to the natural language questions,and better meet the needs of modern society.The research of intelligent question answering system based on the public data of Chinese domain knowledge graph provided by NLPCCICCPOL 2018 and the data of basic education knowledge graph provided by Tsinghua University,and this system is divided into three sub-tasks:entity identification,attribute extraction and system construction.This paper focuses on the three sub-tasks as follows:(1)A multi-modal MCS＿ERNIE+Bi＿LSTM+CRF entity recognition model based on Multi-Character-Sentence(MCS)vector feature fusion is proposed.Firstly,we use ERNIE’s pre-trained language model to learn the relation of natural language questions and carry out vector mapping.Then feature fusion of word level vector and sentence level vector is carried out.Then it is introduced into the Bi＿LSTM model to further extract the features of the entities in the questions.Finally,CRF was used for sequence tagging to identify the entities most related to natural language questions as the entity part of the answer selection.The experimental results show that the F1 value of the entity recognition model reaches 98.62%.(2)An ERNIE+Att＿Bi＿LSTM attribute extraction model based on Attention mechanism is proposed.Firstly,the related entities are obtained from the entity recognition model,and their related attributes are found through the knowledge graph,which are used as the candidate attribute set of natural language questions.Secondly,ERNIE’s pre-trained language model is used to carry out word vector mapping for natural language questions and candidate attribute sets.Then,the Att＿Bi＿LSTM model based on attention mechanism is used to further study the features of natural language questions and candidate attribute sets.Finally,the attributes most related to natural language questions are selected as the attribute part of the answer selection.Experimental results show that the accuracy of this attribute extraction model reaches 95.75%.(3)Through the above two models,the entities related to natural language questions and the attributes representing semantic information of natural language questions are firstly obtained.Firstly,entities related to natural language questions and attributes representing semantic information of natural language questions are obtained through the above two models;Then the answers of natural language questions are queried according to the characteristics of knowledge graph.Finally,the whole process of answering natural language questions is integrated into a complete system,and the use of Python Web lightweight framework Flask and related front-end technology to build a visual interface to interact with users.