Obligatory Relation Detection And Open-world Completion For Knowledge Graphs

In recent years,the research of knowledge graphs(KG)has been widely concerned.The construction and application of large-scale knowledge graphs have greatly promoted the ap-plication of artificial intelligence in various industries and it was called "the advanced ladder of artificial intelligence".However,the quality of large-scale knowledge graphs driven by big data is often not high,and the completeness of the knowledge graphs is an important fac-tor that affects the quality of knowledge graphs.Incomplete knowledge graphs will greatly reduce the performance and user experience of upper-layer applications.In order to improve the completeness of the knowledge graphs,KG engineers often make efforts on the following two aspects:On the one hand,it is to find a way to deter-mine the obligatory relations and attributes information of each entity under the concept in the knowledge graphs,so as to grasp the specific missing information of each entity,and then make targeted completion of the missing information.On the other hand,the missing relationships among entities in knowledge graphs can be directly completed by reasoning through knowledge reasoning and other technical methods.However,the existing methods based on statistical analysis only focus on the distribution of each relation in the knowl-edge graphs under the corresponding concept,which has poor effects on the judgment of the obligatory relations with a high degree of missing.The methods of knowledge graph reasoning completion are divided into closed-world knowledge graph completion(reasoning completion based on the knowledge within KG)and open-world knowledge graph comple-tion(reasoning completion using knowledge outside the KG).The closed-world knowledge graph completion approaches can not introduce new entities and new relations,but the ex-isting open-world knowledge graph completion still has many shortcomings.In order to better complete the missing information of the knowledge graphs,this paper researches and implements a new method of determining the obligatory relation of knowl-edge graphs and a new open-world knowledge graph completion model.Specifically:(1)We propose a new method to determine the obligatory relation of knowledge graphs based on the correlation aggregation model.This method aggregates the related relations of concepts,and can represent and reason the concepts and relations in the same space,thereby inferring the corresponding obligatory relations for each concept(2)We propose a new open-world knowledge graph completion method based on the Multiple Interaction Attention(MIA)model for the external text of knowledge graphs.Com-pared with the current state-of-the-art methods,this model can make full use of the rich feature information in the text description,and will not miss the crucial target information.We performed experiments on several benchmark datasets.The experimental results show that our proposed method for determining the obligatory relations of knowledge graphs has improved the F1 score by more than 7%compared with the state-of-the-art methods,and our open-world knowledge graph completion model also achieves significant improvements compared to state-of-the-art methods.