Research And Prototype Implementation Of Intelligent Question Answering System Based On Temporal Knowledge Graph

The intelligent question answering system based on knowledge graph provides people with more convenient and accurate information services.However,time information plays an indispensable role in portraying the facts.Although temporal knowledge graph contains time information,the research on it is still in its infancy.This thesis is mainly to implement an intelligent question answering system based on temporal knowledge graph.The implementation of this system requires the construction of a temporal knowledge graph and a corresponding question and answer model.Temporal knowledge map provides knowledge source for the whole system,and the question answering model gives the corresponding answers after processing the input questions.In order to realize the above question answering system,this thesis mainly studies the following three aspects:(1)In this thesis,a temporal knowledge extraction model T-CASREL is constructed to complete the construction of temporal knowledge graph.The construction of temporal knowledge graph needs to extract time information from text,but the existing knowledge extraction models only focus on the extraction of entities and relationships,and cannot complete the extraction of temporal information.T-CASREL model uses a special time extraction layer to extract time information,and integrates time information into the corresponding entity and relation information to form a quad,which solves the problem of acquiring time dimension in the temporal knowledge graph.In terms of entity and relation extraction,T-CASREL model adopts pre-training model and joint extraction model,and its advantages in performance are verified through comparative experiments.(2)This thesis constructs TTrans KGQA to implement the question searching function of the question answering system.The performance of traditional question answering model is poor for questions involving time information and multiple knowledge inference.Although the latest question answering model based on knowledge graph embedding achieves good results on multiple knowledge reasoning problems,it cannot be used to answer questions related to time information.In this thesis,the existing knowledge graph embedding model is firstly extended to the time dimension.it can embed the temporal knowledge graph.Then,the extended embedding model is applied to the TTrans KGQA model,and the ability of the model in reasoning problems involving time information and multiple knowledge is verified through experiment.Finally,different knowledge graph embedding models in TTrans KGQA are compared and the best model is selected for the realization of question answering model.(3)In this thesis,an intelligent Q?A system based on temporal knowledge graph constructed by T-CASREL model and TTrans KGQA model is constructed.The system has realized four functions of question search,recent question,knowledge graph visualization and knowledge addition.Each function has passed the test and can meet the needs of users...