The research on knowledge graph completion based on knowledge reasoning has obtained obvious effect on static graph,but it has shortcomings in dealing with time-related events.Knowledge graph completion based on temporal reasoning is more suitable for real events and has higher research value.However,most of the existing temporal knowledge graph completion techniques have limita-tions in processing node information and global information.Therefore,an improved method was pro-posed,which aggregated neighbor information by attention and obtains global time information by using bidirectional LSTM,and completed the missing information in temporal knowledge graph by reasoning prediction.At the same time,the reasoning path graph was generated in the reasoning process to solve the problem of unexplainability caused by neural network.The experimental results,which used 4 different time spans of public data sets and compared with mainstream methods,show that the proposed method is superior to the existing methods in terms of Rmr,h@1 andh@10 indicators.
维普
万方数据