Knowledge Graph Construction And Event Reasoning For COVID-19

The highly contagious and invisible COVID-19 plagued the whole world since its outbreak in Wuhan in 2019.The virus has caused unmeasurable damages although it is soon put under control in China.The major prevention strategy is to study the trajectory data of COVID-19 patients,have their close contacts tested,and prevent further transmission.This paper models various relations hidden in the trajectory data of the patients by using the knowledge graph technology,and analyzes the transmission patterns of the COVID-19 in specific regions based on the visualized results.Based on the COVID-19 temporal knowledge graph constructed from patient trajectory data,this paper proposes a temporal knowledge graph reasoning model,which is used to predict possible epidemic events in future and to send early warnings.This research covers the following work:Firstly,extract key information from the trajectory data of COVID-19 patients announced by the government and point of interest（POI）data,process the information into structured data.The structured data will be used for the COVID-19 knowledge graph visualization and the COVID-19 temporal knowledge graph event reasoning model.Three real datasets were collected: Jilin Province dataset,Heihe City dataset and Dalian City dataset.Taking the Dalian City dataset as an example,this research analyzes the transmission pattern and influencers of the pandemic in Dalian based on the mathematical statistics of gender and age.Secondly,the Neo4 j graph database and Neovis.js tool are used to implement the visualized query system of COVID-19 knowledge graph.In the visualized query system,this paper constructs three knowledge graphs based on varied relations: social relationship knowledge graph,location-level knowledge graph,and COVID-19 knowledge graph.The impact of varied relations on COVID-19 transmission is analyzed based on the visualized results.Thirdly,this research proposes a graph representation learning based temporal knowledge graph reasoning model,namely GRL-TeR,which is used to predict epidemic events based on the COVID-19 temporal knowledge graph.The model is composed of a representation learning module（encoder）and an event reasoning module（decoder）.The temporal knowledge graph is constructed as a historical knowledge graphs sequence with time-stamps.The representation learning module is composed of the Relation Graph Convolution Network（R-GCN）and the Gate Recurrent Unit（GRU）,which encode the structure information and sequence patterns of the historical knowledge graphs sequence into the entity embedding and the relation embedding.The event reasoning module defines the target issue as the entity predicting task and the relation predicting task.With a multi-task scoring function,the conditional probability vector of all entities and relations for a given query is calculated.Lastly,experiments on three datasets demonstrate the advantages of the GRL-TeR model.In the comparison experiment,the advanced static knowledge graph reasoning model and temporal knowledge graph reasoning model are compared,with the superiority of the GRL-TeR model being verified.In the ablation experiment and the model performance analysis experiment,the necessity of all modules in the GRL-TeR model and the rationality of the model experiment are verified.