This paper intends to predict the missing links in temporal knowledge graph for addressing the incompleteness of temporal knowledge graph.The study consists of improving the RE-GCN model based on the existing temporal knowledge graph prediction model;considering all facts related to the enti-ties fully in the process of structural dependence learning in the model evolution unit;extending all the historical facts and enriching the structural dependence relationship between entities based on the consid-eration of the facts adjacent to entities.The results show that compared with the original model,the MMR of entity prediction task and relational prediction task in ICEWS18 dataset increase by 2.02%and 5.58%,respectively.
维普
万方数据