Research Of Question Answering System Based On Temporal Knowledge Graph

With the development of the Internet,obtaining information quickly and accurately from massive data has become an urgent need for people.The knowledge graph based question answering system provides users with convenient personalized information services through structured data content and efficient retrieval methods.Conventional knowledge graphs are usually static and cannot depict the dynamic evolution of facts,which limits their application scenarios.The temporal knowledge graph extends the static knowledge graph in the temporal dimension,including temporal information of knowledge,and has important research value.At present,temporal knowledge graph has become a research hotspot in the field of knowledge graph,but the study of using temporal knowledge graph as the knowledge source of question answering systems is still in its infancy.In order to fully leverage the advantages of modeling structured knowledge dynamic changes using temporal knowledge graphs and improve the coverage and accuracy of question answering systems,this article studies question answering systems based on temporal knowledge graphs,mainly including the following three aspects:(1)Aiming at the shortcomings of existing knowledge extraction models,a knowledge extraction model based on the improved BERT-Bi LSTM-CRF method is proposed to build a temporal knowledge map.Introducing a time extraction module beyond the acquisition of entity and relationship data has achieved the extraction of time dimension information.The fusion of entity,relationship,and corresponding temporal information forms a four tuple temporal knowledge graph.The effectiveness of the temporal knowledge extraction method proposed in this paper is proved by comparative experiments.(2)We have constructed a financial domain temporal knowledge graph dataset and a question and answer dataset.In order to solve the problem of lack of question and answer data set of time series knowledge map,the time series knowledge extraction model in(1)is used to mine the open Chinese text data in the financial field,so as to build a Chinese financial time series knowledge map.On the basis of this knowledge graph,a rule template based method was used to generate a temporal knowledge graph question answering dataset,providing a data foundation for the construction of a question answering system.(3)A question answering system based on temporal knowledge graph named TempKGQA was proposed.This article adopts a temporal knowledge graph representation learning method based on time graph convolutional networks to obtain the representation vectors of entities,relationships,and time,and applies them to question answering systems to achieve problem reasoning that includes time information and multiple pieces of knowledge.Through comparative experiments,it has been proven that the model proposed in this article achieves better results in the Chinese financial domain temporal knowledge graph Q&A compared to existing baseline methods.